Efficient Calibration to FX Options by Markovian Projection

in Cross-Currency LIBOR Market Models

A. Antonov∗ and T. Misirpashaev†

October 10, 2006

Abstract

We revisit the cross-currency LIBOR Market Model armed with the technique of Markovian
projection. We derive an efficient approximation for FX options and show how the FX skew
can be modeled consistently with the interest rate skew in a common multifactor model.

Introduction
Extension of the LIBOR Market Model (LMM) framework to cross-currency instruments is
conceptually transparent and was addressed by Mikkelsen (2001) and Schlögl (2002) at a time
of widespread acceptance of the benefits of single-currency LMMs in the valuation of fixed-
income exotics. A parallel improvement of the tools for simulation-based treatments of the
early exercise (see e.g., recent reviews by Piterbarg (2004) and Fries (2005)) alleviated the
associated computational difficulties and led to a displacement of short-rate models, especially
those with one factor, from the position of choice in single-currency applications. However, with
regard to cross-currency exotics such as PRDCs, the mainstream approach is still to employ
one-factor short-rate models for the interest rate components. A particularly tractable and
therefore popular model arises when both interest rates are described by normal models (such
as the Hull-White model) while the FX rate is described by the log-normal Black-Scholes model.
The calibration to long-dated European FX options in this model is fully analytical, but the
skew of implied volatilities for single-currency options and FX options cannot be captured.
In a recent work, Piterbarg (2006a) advanced a local volatility model for the FX component
and developed efficient methods for the calibration to FX skew. The interest rate components
remained short-rate Gaussian. In this work we start with a skewless log-normal process for the
spot exchange rate but with skewed—also known as shifted log-normal or displaced-diffusion—
LMM interest rate components. This approach allows for a consistent treatment of interest
rate skew and consistent pricing of single-currency and cross-currency instruments within a
common multifactor model. A notable technical difficulty which has prevented a wider roll-out
∗
NumeriX Software Ltd. 41 Eastcheap Street, 2nd Floor London EC3M 1DT UK; [email protected]
†
NumeriX LLC 4320 Winfield Rd, Suite 200, Warrenville, IL 60555 USA; [email protected]

1
of cross-currency LMMs is the simultaneous calibration to long-dated European FX options at
several maturities. Despite possible optimizations (Amin (2003)), a simulation-based calibration
remains insufficient. We tackle the problem of efficient analytical approximation using the
methods of Markovian projection and parameter averaging in the footsteps of Piterbarg (2006b).
We begin by setting up the cross-currency LIBOR market model in the spot-LIBOR measure.
Then we proceed to the description of the calibration procedure and the derivation of the
analytical approximation for the FX option, followed by numerical results and conclusions. Key
technical steps of the derivation are outlined in the Appendix.

Cross-currency LMM setup

All components of the cross-currency model are defined on the same probability space equipped
with a filtration Ft which is a suitable augmentation of the filtration generated by a standard F -
dimensional Brownian motion Z(t). We assume a common set of maturities 0 = T0 < T1 < . . . <
TN for both single-currency rate LMM components, and define domestic and foreign LIBORs,
Ln (t) and L̃n (t), as forward rates starting at Tn and ending at Tn+1 , generally reserving the
tilde for foreign-currency quantities. In terms of T -maturity zero-coupon bonds, P (t, T ) and
P̃ (t, T ), the LIBORs are given by
µ ¶ Ã !
1 P (t, Tn ) 1 P̃ (t, Tn )
Ln (t) = −1 , L̃n (t) = −1 , (1)
δn P (t, Tn+1 ) δn P̃ (t, Tn+1 )
where δn is the daycount fraction from Tn to Tn+1 .
Stochastic evolution of the LIBORs is determined by the volatility functions γn (t) and γ̃n (t),
assumed deterministic. Each volatility function and each Brownian motion have F components
by the number of factors, but the factor index is always omitted. Where no confusion is possible,
we will write scalar products in the space of factors simply as f g (such as γn (t)dZn (t) in the
formula below). In more complicated cases, we will use the notation (f ·g).
We assume a displaced diffusion for the evolution of each LIBOR in its martingale measure,
dLn (t) = (bn (t)Ln (t) + (1 − bn (t))Ln (0))γn (t)dZn (t),
dL̃n (t) = (b̃n (t)L̃n (t) + (1 − b̃n (t))L̃n (0))γ̃n (t)dZ̃n (t) (2)
(bn ≡ 1 corresponds to the purely log-normal case while bn ≡ 0 is the normal limit). Thus we
start with a skew model for each single-currency market. The calibration of the time-dependent
shifts bn (t), b̃n (t) can be efficiently done using Piterbarg’s averaging formula quoted in the next
section as Eq. (19).
An important part in the analytics will be played by bond ratios,
P (t, Tn )
Rn (t) = = 1 + δn Ln (t),
P (t, Tn+1 )
P̃ (t, Tn )
R̃n (t) = = 1 + δn L̃n (t). (3)
P̃ (t, Tn+1 )
The SDEs for the bond ratios in the martingale measure have the same form as those for the
LIBORs,
dRn (t) = (βn (t)Rn (t) + (1 − βn (t))Rn (0))σn (t)dZn (t),
dR̃n (t) = (β̃n (t)R̃n (t) + (1 − β̃n (t))R̃n (0))σ̃n (t)dZ̃n (t), (4)

2
with
bn (t)Rn (0) Rn (0) − 1
βn (t) = , σn (t) = γn (t) . (5)
Rn (0) − 1 Rn (0)
We take for the numeraire in the domestic currency the rolling spot account, corresponding
to Jamshidian’s (1997) spot-LIBOR measure,
1 1 P (t, Tn+1 )
N (t) = ··· for Tn < t ≤ Tn+1 , (6)
P (T0 , T1 ) P (Tn−1 , Tn ) P (Tn , Tn+1 )
and set the volatility of the short bond P (t, Tn+1 ) to 0. As was shown by Schlögl (2002),
this numeraire is equivalent to the continuously compounded savings account in the sense that
they lead to the same measure. Another property shared by N (t) and the savings account is
diffusion-free dynamics,
dN (t) = N (t) µN (t)dt. (7)
The foreign-currency numeraire Ñ (t) is chosen similarly.
The FX dynamics linking the single-currency models into a cross-currency model are defined
by postulating log-normal dynamics for a numeraire-discounted asset Y (t) related to the foreign
exchange rate X(t) by
X(t) Ñ (t)
Y (t) = . (8)
N (t)
We set
dY (t) = Y (t)σY (t)dZ(t) (9)
in the spot-LIBOR measure with a deterministic F -component volatility, σY . The dynamics of
the FX rate X(t) follow from Eqs. (8) and (9) and the deterministic character of the numeraire
dynamics given by Eq. (7),

dX(t)/X(t) = (µN (t) − µ̃N (t)) dt + σY (t)dZ(t). (10)

Note that we have not introduced any skew in the process for Y (t). We will see that the FX
options skew will be naturally generated by the interest rate components even if we started with
purely log-normal LIBOR dynamics.
We complete the operational definition of the model by restating the dynamics of the LIBORs
in the domestic spot-LIBOR measure,
 
n
X
dLn (t) = (Ln (t) bn (t) + Ln (0) (1 − bn (t)))γn (t)  αk (t) dt + dZ(t) ,
k=η(t)
 
n
X
dL̃n (t) = (L̃n (t) b̃n (t) + L̃n (0) (1 − b̃n (t)))γ̃n (t)  α̃k (t) dt − σY dt + dZ(t) . (11)
k=η(t)

Here η(t) = n + 1 for Tn < t ≤ Tn+1 , and

bk (t)Lk (t) + (1 − bk (t))Lk (0)
αk (t) = δn γk (t),
1 + δk Lk (t)
b̃k (t)L̃k (t) + (1 − b̃k (t))L̃k (0)
α̃k (t) = δn γ̃k (t). (12)
1 + δk L̃k (t)

3
Note that the drift terms in this equation are of the second order in volatilities.
The model can be simulated by inductively repeating the following steps after initialization
at the origin of time.
• Extend the paths of the F random factors to the next date.
• Get domestic and foreign states for the next date from a discretization of Eqs. (11).
• Get domestic and foreign numeraires, N (t) and Ñ (t), from Eq. (6) and its foreign-currency
counterpart.
• Get the value of the process Y (t) for the next date from a discretization of Eq. (9).
• Restore the FX rate at the next date as X(t) = Y (t)N (t)/Ñ (t).
The valuation of early-exercise options embedded in the exotic instruments can be done using
the techniques reviewed by Piterbarg (2004) and Fries (2005). In the next section, we turn to
the model calibration.

Calibration of FX volatility
The calibration of the volatilities γn (t) and γ̃n (t)—or, equivalently, σn (t) and σ̃n (t)—to single-
currency caps and swaptions, and correlations between single-currency rates, can be done in-
dependently for the domestic and foreign components using the techniques developed for the
single-currency LMM. The overlap in the factor structure of volatilities can be calibrated to
cross-correlations of foreign and domestic rates in a straightforward way. Here, we address the
key issue of the calibration to long-dated European FX options. Schlögl (2002) studied the
linkage between domestic and foreign rate measures and pointed out that, with a choice of
log-normal LIBOR models in both domestic and foreign markets, a log-normal distribution was
possible for the forward FX rates at one maturity only, which precluded efficient calibration.
Here we dispense with the log-normality desideratum for the forward FX rate at any maturity,
and instead develop an efficient analytical approximation valid at all maturities.
To compute the option price with maturity TM and strike K, we change the measure to the
TM -forward measure having as a numeraire the domestic bond P (t, TM ). The forward FX rate

X(t) P̃ (t, TM )
F (t, TM ) = (13)
P (t, TM )

is a martingale under the TM -forward measure. The option price reduces to an expectation
under this measure,

(X(TM ) − K)+
· ¸
= P (0, TM )ETM (F (TM , TM ) − K)+ .
£ ¤
E (14)
N (TM )

Using Eqs. (3), (6), (8), and (13), the forward FX-rate process can be expressed in terms of
the bond ratios, Rn (t) and R̃n (t), and the process Y (t),
η(t)−1 M −1
Y P̃ (Tk , Tk+1 ) Y
F (t, TM ) = Y (t) Rn (t)R̃n−1 (t). (15)
P (Tk , Tk+1 )
k=0 n=η(t)

4
 Below we assume that the maturity TM of the forward FX rate is fixed and omit the corre-
sponding argument, writing F (t) instead of F (t, TM ). An SDE for F (t) can be obtained by Ito’s
lemma from Eq. (15) using the diffusion terms from Eqs. (4) and (9). Note that we know with-
out calculation that the drift terms will cancel because we now work in a martingale measure
for F (t). We get
dF (t) = F (t)Λ(t)dW (t), (16)

Λ(t) = σY (t)
M −1
X βn (t) Rn (t) + (1 − βn (t)) Rn (0)
+ σn (t)
Rn (t)
n=η(t)
M −1
X β̃n (t) R̃n (t) + (1 − β̃n (t)) R̃n (0)
− σ̃n (t). (17)
n=η(t)
R̃n (t)

The forward-rate process is obviously non-Markovian. Our strategy is to replace it with an

effective Markovian process F ∗ (t) following a time-dependent displaced diffusion

dF ∗ (t) = (F ∗ (t)β(t) + (1 − β(t))F ∗ (0)) σ(t)dW (t), F ∗ (0) = F (0), (18)

such that its marginal distributions well approximate those of the initial true process F (t). Once
this is done, we invoke the formula for shift averaging derived by Piterbarg (2005),
R TM 2 (t) t σ 2 (τ )dτ dt
R
β(t)σ
β̄TM = 0R T Rt 0 , (19)
M 2 2
0 σ (t) 0 σ (τ )dτ dt

which reduces the pricing of an FX option with maturity TM to a simple modification of the
Black-Scholes formula.
We now solve the problem of finding the optimal shift β(t) and volatility σ(t) in Eq. (18)
using an elaboration on Piterbarg’s (2006) technique of Markovian projection. The technique
is based on a theorem by Gyöngy (1986) and Dupire (1997) which asserts that the process

dF ∗∗ (t) = Σ(t, F ∗∗ (t))dW (t), F ∗∗ (0) = F (0) (20)

has exactly the same marginal distributions as F (t) provided

h ¯ i
2 2 2¯
|Σ(t, x)| = E F (t)|Λ(t)| ¯ F (t) = x . (21)

For every fixed t, the conditional expectation in the right hand side can be characterized as a
function of state |Σ(·, x)|2 which minimizes the L2 -distance from the true variance,
h¡ ¢2 i
χ2 = E F 2 (t)|Λ(t)|2 − |Σ(t, F (t))|2 → min. (22)

We minimize the functional in Eq. (22) over a subspace of affine linear functions of state,
expressed in the form
Σ(t) = (F (t)β(t) + (1 − β(t))F (0)) σ(t). (23)

5
This corresponds to a minimization over β(t) and σ(t) of a function
·³ ´2 ¸
2 2 2 2 2
χL (σ(t), β(t)) = E F (t)|Λ(t)| − (F (0) + β(t)∆F (t)) |σ(t)| (24)

for every fixed t. Here, ∆F (t) = F (t) − F (0). Equating to zero the variations of χ2L (σ(t), β(t))
over σ(t) and β(t) gives a pair of equations,

E F 2 (t)|Λ(t)|2 (F (0) + β(t)∆F (t))2 = |σ(t)|2 E (F (0) + β(t)∆F (t))4 ,

£ ¤ £ ¤
(25)
£ 2 2
¤ 2
£ 3
¤
E F (t)|Λ(t)| (F (0) + β(t)∆F (t))∆F (t) = |σ(t)| E (F (0) + β(t)∆F (t)) ∆F (t) . (26)

Solving these in the lowest leading order in volatilities σY (t), σn (t), σ̃n (t) gives the result
M
X −1
σ(t) = σY (t) + (σn (t) − σ̃n (t)) , (27)
n=η(t)

MP−1 Rt
(1 − βn (t))(σn (t)·σ(t)) 0 (σn (τ )·σ(τ ))dτ
n=η(t)
β(t) = 1 − Rt
|σ(t)|2 |σ(τ )|2 dτ
0
MP
−1 Rt
(1 − β̃n (t))(σ̃n (t)·σ(t)) 0 (σ̃n (τ )·σ(τ ))dτ
n=η(t)
+ Rt . (28)
|σ(t)|2 2
0 |σ(τ )| dτ

Additional technical details of the derivation can be found in the Appendix.

Recall that the skew parameter β(t) identically equal to 1 corresponds to a purely log-normal
distribution and the absence of implied volatility skew. Eq. (28) produces no skew in the FX
option if there is no skew in all bond ratios, βn (t) = β̃n (t) ≡ 1. This, however, would not be
the case if we started with log-normal LIBORs, bn (t) = b̃n (t) ≡ 1, as can be seen from Eq. (5).
The pricing of the European FX option (14) with maturity TM and strike K finally reduces
to a variant of the Black-Scholes formula,

(X(TM ) − K)+
· ¸ µ µ ¶ ¶
F F (1 − β̄TM )
E = P (0, TM ) N (d+ ) − K + N (d− ) , (29)
N (TM ) β̄TM β̄TM
TM
ln F/(K β̄TM + F (1 − β̄TM )) ± V /2
Z
d± = √ , V = β̄T2M |σ(t)|2 dt, (30)
V 0

where F = F (0, TM ) is the forward rate, and β̄TM is the effective shift obtained by applying the
averaging formula (19) to the time-dependent shift (28).

Numerical results
In this section we check numerically our approximation for the European FX option pricing.
We used recent data for swap rates (Fig. 1) and ATM caplet volatilities in EUR (domestic)
and USD (foreign) markets, and for long-dated European EUR/USD options to run a test

6
 with a 3-factor log-normal LIBOR market model. The absolute value of the instantaneous
FX volatility |σY (t)| is plotted in Fig. 2 as a function of time. The absolute values of time-
independent LIBOR volatilities |γn | and |γ̃n | are plotted in Fig. 3 as functions of semi-annual
LIBOR maturity Tn . Each volatility has three components with the weights chosen to obtain
specific time-independent values of effective correlations,

(σY ·γn ) (σY ·γ̃n ) (γn ·γ̃n )

= −15%, = −20%, = 25%. (31)
|σY ||γn | |σY ||γ̃n | |γn ||γ̃n |

Table 1 shows the values of maturities and strikes of FX options used in the test. Ta-
ble 2 shows implied BS volatilities computed analytically using the approximation based on the
Markovian projection formulas (27, 28) and the parameter-averaging formula (19). The errors
of the approximation with respect to a simulation with a large number of paths are in Table 3.
Implied volatility skews plotted in Figs. 4–6 show an excellent ability of the analytics to
follow the exact pattern of skew computed using a simulation with a large number of paths.
We observe a small systematic bias in the absolute level of the implied volatility curve, the
correction of which would require taking higher-order corrections to the effective volatility (27)
into account. Note that a noticeable skew of FX options is obtained starting from purely log-
normal models for the LIBORs and without assuming any explicit skew in the process of the
FX rate, which suggests that consistent log-normal approximations for FX options are neither
possible nor desirable.

Swap rates

EUR (domestic) USD (foreign)

6.0%

5.5%

5.0%

4.5%

4.0%

3.5%

3.0%
0 5 10 15 20 25 30
Swap maturity in years

Figure 1: Swap rate curves in EUR (domestic) and USD (foreign) markets used in the test.

7
 EUR/USD instantaneous volatility

25.0%

20.0%

15.0%

10.0%

5.0%

0.0%
0 5 10 15 20 25 30
Time in years

Figure 2: Absolute value of instantaneous volatility |σY (t)| of EUR/USD rate used in the test.

Model volatilities of Libors

EUR (domestic) USD (foreign)

18.0%
17.0%
16.0%
15.0%
14.0%
13.0%
12.0%
11.0%
10.0%
0 5 10 15 20 25 30
Libor maturity in years

Figure 3: Absolute values of the LMM volatilities for domestic and foreign markets, |γn (0)| and
|γ̃n (0)|, calibrated to interest rate caps in the corresponding currencies and plotted as a function of
Tn . The volatilities are chosen to be time-independent, γn (t) ≡ γn (0).

8
 Maturity set 1 set 2 set 3 set 4 set 5 set 6 set 7
5Y 66.77 74.67 83.50 93.38 104.43 116.78 130.59
10Y 54.70 64.07 75.04 87.90 102.95 120.59 141.25
15Y 46.72 56.71 68.82 83.53 101.37 123.04 149.32
20Y 40.22 50.30 62.90 78.66 98.37 123.03 153.85
25Y 35.26 45.27 58.13 74.64 95.84 123.06 158.01
30Y 30.96 40.71 53.53 70.40 92.58 121.74 160.09

Table 1: Maturities and strikes of European FX options used in the numerical test. Strikes are
expressed in % of spot. The column with values in bold face corresponds to ATM strikes.

Maturity set 1 set 2 set 3 set 4 set 5 set 6 set 7

5Y 9.43 9.40 9.38 9.36 9.34 9.32 9.31
10Y 11.03 10.90 10.78 10.68 10.58 10.50 10.42
15Y 13.65 13.43 13.24 13.07 12.92 12.78 12.66
20Y 16.47 16.19 15.95 15.74 15.55 15.39 15.26
25Y 18.95 18.60 18.31 18.07 17.85 17.67 17.52
30Y 21.25 20.84 20.50 20.21 19.97 19.77 19.60

Table 2: Implied volatilities of European FX options computed using the analytical approximation
based on Markovian projection.

Maturity set 1 set 2 set 3 set 4 set 5 set 6 set 7

5Y -1 -1 -1 -1 -1 -1 -1
10Y -8 -6 -5 -4 -5 -6 -7
15Y -10 -8 -6 -6 -7 -9 -13
20Y -7 -7 -6 -7 -9 -11 -15
25Y -1 -2 -4 -6 -9 -12 -16
30Y 11 5 1 -4 -7 -10 -15

Table 3: Basis point errors of the analytical approximation expressed in terms of deviations in the
the values of implied BS volatilities.

9
 Implied BS volatility, maturity 5y

9.50

9.45

9.40 analytical
log-normal
9.35 simulation

9.30

9.25
60 80 100 120 140
Figure 4: Comparison of analytics and simulation for implied BS volatility skew at 5y.

10
 Implied BS volatility, maturity 15Y

14.00

13.80

13.60

13.40 analytical
log-normal
13.20 simulation

13.00

12.80

12.60
40 60 80 100 120 140 160
Figure 5: Comparison of analytics and simulation for implied BS volatility skew at 15y.

11
 Implied BS volatility, maturity 30Y

22.00

21.50

21.00
analytical
20.50 log-normal
simulation
20.00

19.50

19.00
20 40 60 80 100 120 140 160 180
Figure 6: Comparison of analytics and simulation for implied BS volatility skew at 30y.

12
Conclusions and future directions
We revisited the setup of cross-currency LIBOR Market Models and showed how the technique
of Markovian projection can be applied to the problem of calibration to FX options in the case
when the presence of skew makes an assumption of log-normal forward FX rates inappropriate.
In doing so, we developed a new approach to the Markovian projection, based on an explicit
analytical minimization of the functional-space distance from the exact stochastic volatility to
the low-dimensional subspace of displaced diffusion volatilities, given by Eq. (22).
We found that the skew in implied volatilities of FX options is present even if it was not
built explicitly into the exchange rate model. The FX skew generated in this way from single-
currency LMMs may not be sufficient to fit the market data. To obtain a more flexible model,
we could start with a displaced-diffusion version of Eq. (9), which is addressed in a separate
work.
We are indebted to Vladimir Piterbarg for an introduction into the method of Markovian
projections. AA acknowledges a stimulating discussion with Peter Jäckel as a result of which
we became aware of a forthcoming unpublished work by Jäckel and Kawai where the problem of
FX skew in cross-currency LMM modeling is addressed using a different set of analytical tools.
We are grateful to our colleagues at NumeriX and especially to Gregory Whitten for supporting
our work.

Appendix: Perturbation theory for effective skew

We assume that an expansion in the powers of volatilities is possible in Eqs. (25) and (26) and
measure the order of every term as a power of the dominant volatility σ+ = max {σn (t), σ̃n (t), σY (t)}.
Keeping only the terms of the order O(σ+ 2 ) in Eq. (25), we get

F (0)2 E[F (t)2 |Λ(t)|2 ] = |σ(t)|2 F (0)4 + O(σ+

4
). (32)

In the leading order, the expected value in the left hand side can be obtained by freezing
the process F (t) → F (0) as well as the bond ratios Rn (t) → Rn (0), R̃n (t) → R̃n (0) in the
definition (17) of Λ(t). This leads to Eq. (27) as one of the solutions (the others are related by
an orthogonal rotation in the factor space and lead to equivalent dynamics).
The lowest order balance in Eq. (26) is of order O(σ+ 4 ),

F (0)E[F (t)2 |Λ2 (t)|2 ∆F (t)] + β(t)E[F 2 |Λ2 (t)|2 ∆F (t)2 ] = (33)
2 2 2 6
3β(t)F (0) |σ(t)| E[∆F (t) ] + O(σ+ ),

which entails
F (0)E[F 2 |Λ2 (t)|2 ∆F (t)] 2
β(t) = + O(σ+ ). (34)
3|σ(t)|2 F (0)2 E[∆F (t)2 ] − E[F 2 |Λ2 (t)|2 ∆F (t)2 ]
The leading order of the averages in this expression cannot be directly found by a replacement of
the processes by their initial values because the result would vanish. A general strategy is to use
Ito’s lemma to take consecutive differentials of the processes appearing under the expectation
sign until the replacement by the initial values becomes possible. For example,

d(∆F (t)2 ) = 2F (t)∆F (t)Λ(t)dW (t) + F (t)2 |Λ(t)2 |dt. (35)

13
Taking the expectations, we obtain an ODE
dE[∆F (t)2 ] = E[F (t)2 |Λ(t)|2 ]dt, (36)
which is easily solved in the leading order,
Z t
2 2
E[∆F (t) ] = F (0) |σ(τ )|2 dτ + O(σ+
4
). (37)
0

This approach also works for a more difficult average

Z t
E[F (t)2 |Λ2 (t)|2 ∆F (t)] = 2|σ(t)|2 |σ(τ )|2 dτ
0
M
X −1 Z t
−2 (1 − βn (t))(σn (t)·σ(t)) (σn (τ )·σ(τ ))dτ
n=η(t) 0

M
X −1 Z t
+2 (1 − β̃n (t))(σ̃n (t)·σ(t)) (σ̃n (τ )·σ(τ ))dτ. (38)
n=η(t) 0

The derivation will be presented elsewhere and can be obtained from the authors on request.
From this, Eq. (28) for effective skew follows.

References
[1] A. Amin (2003) “Multi-factor Cross Currency LIBOR Market Models: Implementation,
Calibration and Examples,” working paper, unpublished.
[2] B. Dupire (1997) “A unified theory of volatility”, Banque Paribas working paper.
[3] C. Fries (2005) “Pricing Options with Early Exercise by Monte Carlo Simulation: an
Introduction,” available at http://www.christian-fries.de/finmath/earlyexercise/
[4] I. Gyöngy (1986) “Mimicking the One-Dimensional Marginal Distributions of Processes
Having an Ito Differential” Probability Theory and Related Fields, 71, 501–516.
[5] F. Jamshidian (1997) “LIBOR and Swap Market Models and Measures”, Finance and
Stochastics, 1, 293–330.
[6] P. Mikkelsen (2001) “Cross-currency LIBOR Market Models,” working paper series No.85,
Center for Analytical Finance Aarhus School of Business.
[7] V. Piterbarg (2004) “Pricing and Hedging Callable Libor Exotics in Forward Libor Models,”
Journal of Computational Finance, 8(2), 65–117.
[8] V. Piterbarg (2005) “Stochastic Volatility Model with Time-dependent Skew,” Applied
Math. Finance, 12, 147–185.
[9] V. Piterbarg (2006a) “Smiling Hybrids,” Risk, May, 66–71.
[10] V. Piterbarg (2006b) “Markovian Projection Method for Volatility Calibration,” available
at SSRN: http://ssrn.com/abstract=906473.
[11] E. Schlögl, (2002) “A Multicurrency Extension of the Lognormal Interest Rate Market
Models,” Finance and Stochastics, Springer, 6(2), pages 173–196.

14