Z
Rd
£βi(t, x, T, T∗) − 1¤ ¡
µL− νTL∗
¢(dt, dx)
− Z
Rd
£βi(t, x, T, T∗) − 1¤ξˆi(t, x, T∗)νTL∗(dt, dx)
(2.2.80)
for 0 ≤ t ≤ T ≤ T∗.
2.2.7 Discrete-tenor L´ evy domestic Libor rate model
Following the approach of Eberlein and ¨Ozkan (see again Eberlein and ¨Ozkan (2002)), we will model forward Libor rates (see Definition 2.1.5) with time-inhomogeneous L´evy processes. Let δ > 0 and T ≤ T∗ − δ be fixed, then we define the domestic Libor rate L(t, T ) by
L(t, T ) = L(0, T ) exp
Zt
0
λ(s, T )>dLT +δs
where
LT +δt = Zt
0
bT +δs ds + Zt
0
csdWsT +δ+ Zt
0
Z
Rd
x¡
µL− νT +δL ¢
(ds, dx)
is a d–dimensional time-inhomogeneous L´evy process such that WT +δ is a PT +δ–standard Brownian motion in Rd, C(t) is a symmetric positive semidefinite d × d matrix, and ct is a measurable version of the square root of Ct.
The random measure of jumps of L is given by µL, and νT +δL is its PT +δ–compensator satisfying
νT +δL ([0, t], G) = Zt
0
Z
Rd
1{[0,t]×G}(s, x)λT +δs (dx) ds, G ∈ B¡
Rd\ {0}¢ ,
where λT +δt is a measure on Rd, which integrates (|x|2∧ 1). Furthermore, for all t ∈ [0, T +δ]
holds λt({0}) = 0.
We assume that for all T ≤ T∗− δ the following integrability conditions hold:
ZT
0
¯¯bT +δs ¯
¯ ds < ∞, ZT
0
kCsk ds < ∞, ZT
0
Z
|x|>1
exp¡ u>x¢
λT +δs (dx) ds < ∞,
where |u| ≤ (1 + ²)MT +δ for some constants MT +δ > 1 and ² > 0. Hence, LT +δt is a special semimartingale with respect to the domestic forward measure PT +δ.
Assumption 2.2.12 The volatility of the domestic forward Libor rate process λ(·, T ) : [0, T ] 7→ Rd+ is a bounded deterministic function such that λ(t, T ) = (0, . . . , 0)> for t > T . Furthermore, there exist positive constants M and M such that M ≤ M and
0 < M ≤ |λ(·, T )| ≤ M < MT +δ for each t ∈ [0, T ].
Observe that by Propositions 2.2.4 and 2.2.5 the volatility function λ(·, T ) is integrable with respect to the time-inhomogeneous L´evy process LT +δt , and the process Rt
0
λ(s, T )>dLT +δs is a special semimartingale, which is also exponentially special under PT +δ.
Furthermore, we assume that the following initial condition is satisfied L(0, T ) = 1
δ
µ B(0, T ) B(0, T + δ) − 1
¶
(2.2.81) for all T ∈ (0, T∗− δ], where the initial term structure B(0, T ), T ∈ (0, T∗] is strictly positive and decreasing in T .
The domestic Libor rate process must be a local martingale under PT +δ. Applying Theo-rem 2.19 in Kallsen and Shiryaev (2002), the martingality of (L(t, T ))0≤t≤T can be assured by choosing the drift term bT +δt such that Rt
0
λ(s, T )>bT +δs ds is equal to the exponential compensator of L(t, T ). Applying Proposition 2.2.5, we get
Zt
0
λ(s, T )>bT +δs ds = − eKtLT +δ(λ(·, T ))
= −1 2
Zt
0
λ(s, T )>Csλ(s, T ) ds
− Zt
0
Z
Rd
£exp¡
λ(s, T )>x¢
− 1 − λ(s, T )>x¤
νT +δL (ds, dx),
where eKtLT +δ(λ(·, T )) denotes the Laplace cumulant process of LT +δ in λ(·, T ).
Hence, we obtain the following dynamics of L(·, T ) under the forward measure PT +δ
dL(t, T ) = L(t−, T )
λ(t, T )>ctdWtT +δ+
+ Z
Rd
¡exp¡
λ(t, T )>x¢
− 1¢ ¡
µL− νT +δL ¢
(dt, dx)
.
(2.2.82)
Recall that FB(t, T, T +δ) = 1+δL(t, T ), therefore we derive from (2.2.82) that the forward process FB(·, T, T + δ) satisfies
dFB(t, T, T + δ)
FB(t−, T, T + δ) = δL(t−, T ) 1 + δL(t−, T )
λ(t, T )>ctdWtT +δ
+ Z
Rd
¡exp¡
λ(t, T )>x¢
− 1¢ ¡
µL− νT +δL ¢
(dt, dx)
.
Observe that the coefficients
δL(t−, T )
1 + δL(t−, T )λ(t, T )>ct, and δL(t−, T )
1 + δL(t−, T )
£exp¡
λ(t, T )>x¢
− 1¤
are not deterministic. Since we want to stay within the class of PIIAC processes, we have to approximate the last equations by replacing the random term
l(t−, T ) := δL(t−, T ) (1 + δL(t−, T ))
with a deterministic one l(0, T ) = δL(0, T )/(1 + δL(0, T )). A similar approximation has been already used in Brace and Womersley (2000) for the derivation of the approximate swaption formula in the lognormal forward Libor model, as well as in Schl¨ogl (2002).
Notice that the quotient
l(t−, T ) = 1 − 1
1 + δL(t−, T ) = 1 − FB(t−, T + δ, T ) ∈ [0; 1].
The suggested approximation can be easily justified, since, in absolute terms, not only is l(t−, T ) bounded in [0; 1], but its absolute volatility is a couple of orders of magnitude less than the volatilities of the forwards. Following the approach of Musiela and Rutkowski (see Musiela and Rutkowski (1997a)), we build an arbitrage-free domestic bond market, which is based on the Libor model presented above, using backward induction procedure.
More specifically, given a discrete-tenor structure T = {T0 < T1 < T2 < TN < TN +1= T∗} with δ = Tj+1− Tj, j = 0, . . . , N, one can construct the forward measure for the date Tj−1 with j = 2, . . . , N + 1, provided that the forward measure PTj, the standard Brownian motion WTj, the compensator νTLj of the jump measure µL, and the volatility λ(·, Tj−1) of the domestic Libor rate L(·, Tj−1), are known. In particular, under the forward measure PTj with j = 1, . . . , N + 1 the domestic Libor rate is defined by
L(t, Tj−1) = L(0, Tj−1) exp
Zt
0
λ(s, Tj−1)>dLTsj
, (2.2.83)
where
LTtj = Zt
0
bTsjds + Zt
0
csdWsTj+ Zt
0
Z
Rd
x
³
µL− νTLj
´
(ds, dx), (2.2.84)
and for all t ≤ Tj−1 holds Zt
0
¯
¯bTsj¯
¯ + kCsk + Z
Rd
¡|x|2∧ 1¢
λTsj(dx)
ds < ∞
for each j = 1, . . . , N + 1. Furthermore, we make the following integrability assumption.
Assumption 2.2.13 One can find such constants M > 1 and ² > 0, that for each j = 1, . . . , (N + 1) holds
TZj−1
0
Z
|x|>1
exp¡ u>x¢
λTsj(dx) ds < ∞, (2.2.85)
for |u| ≤ (1 + ²)MTj with MTj := M − (N − j + 1) fM and M − (N + 1) fM > max{1, M}, where we have set fM := M/(1 + ²) and M := MTN +1 = MT∗.
Note that under Assumption 2.2.13, we have XN
k=1
λ(t, Tk) ≤ XN k=1
|λ(t, Tk)| ≤ NM < M
for all t ∈ [0, TN]. Moreover, with respect to Proposition 2.2.4 the volatility function λ(·, Tj−1) is integrable with respect to LTtj and the process Rt
0
λ(s, Tj−1)>dLTsj is a special semimartingale, which is also exponentially special.
To guarantee that L(·, Tj−1) is a local martingale under domestic forward measure PTj, we choose the drift term bTj such that Rt
0
λ(s, Tj−1)>bTsjds is equal to the exponen-tial compensator of Rt
0
λ(s, Tj−1)>dLTsj, where t ∈ (0, Tj−1] and j = 1, . . . , N + 1. Using Proposition 2.2.5, we obtain
Zt
0
λ(s, Tj−1)>bTsjds = − eKtLTj(λ(·, Tj−1))
= −1 2
Zt
0
λ(s, Tj−1)>Csλ(s, Tj−1) ds
− Zt
0
Z
Rd
£exp¡
λ(s, Tj−1)>x¢
− 1 − λ(s, Tj−1)>x¤
νTLj(ds, dx).
(2.2.86) With this specification of a drift term, the Libor rate L(t, Tj−1) can be represented as a stochastic exponential of the form
L(t, Tj−1) = L(0, Tj−1) E
Z·
0
λ(s, Tj−1)>csdWsTj
+ Z·
0
Z
Rd
¡exp¡
λ(s, Tj−1)>x¢
− 1¢ ³
µL− νTLj
´
(ds, dx)
t
.
Hence, domestic forward process admits the following representation:
FB(t, Tj−1, Tj) = FB(0, Tj−1, Tj) E¡
HTj−1¢
t, where
HtTj−1 = Zt
0
l(s−, Tj−1)λ(s, Tj−1)>csdWsTj
+ Zt
0
Z
Rd
l(s−, Tj−1)¡ exp¡
λ(s, Tj−1)>x¢
− 1¢ ³
µL− νTLj´
(ds, dx).
To simplify the notation, we put
α (t, Tj−1, Tj)> := l(t−, Tj−1)λ(t, Tj−1)>ct
≈ l(0, Tj−1)λ(t, Tj−1)>ct, β (t, x, Tj−1, Tj) := l(t−, Tj−1)¡
exp¡
λ(t, Tj−1)>x¢
− 1¢ + 1
≈ l(0, Tj−1)¡ exp¡
λ(t, Tj−1)>x¢
− 1¢ + 1.
(2.2.87)
The domestic forward martingale measure associated with the date Tj−1can be defined by setting its Radon–Nikod´ym derivative
dPTj−1
dPTj = FB(Tj−1, Tj−1, Tj)
FB(0, Tj−1, Tj) = 1 + δL(Tj−1, Tj−1)
1 + δL(0, Tj−1) PTj–a.s.
for j = 2, . . . , N + 1. Then a standard Brownian motion and the PTj−1–compensator of the jump measure µL are defined by
WtTj−1 = WtTj− Zt
0
α (s, Tj−1, Tj) ds,
νTLj−1(dt, dx) = β (t, x, Tj−1, Tj) νTLj(dt, dx).
Observe that there is a following relationship between the driving time-inhomogeneous L´evy processes with respect to the forward measures PTj and PTj−1
LTtj = LTtj−1 + Dt(Tj−1, Tj) , where the process Dt(Tj−1, Tj) is given by
Dt(Tj−1, Tj) = Zt
0
¡bTsj − bTsj−1¢ ds +
Zt
0
csα(s, Tj−1, Tj) ds
+ Zt
0
Z
Rd
x (β (s, x, Tj−1, Tj) − 1) νTLj(ds, dx).
(2.2.88)
Note also that for j = 1, ..., N we can establish the following connection between the domestic forward measures PTj and PT∗
WtTj = WtT∗− Zt
0
α (s, Tj, T∗) ds,
νTLj(dt, dx) = β (t, x, Tj, T∗) νTL∗(dt, dx),
(2.2.89)
where We can avoid approximations in equations (2.2.87) and (2.2.90) by modelling the do-mestic forward rate FB(t, Tj−1, Tj) = 1 + δL(t, Tj−1) (instead of Libor rate L(t, Tj−1)) for j = 1, . . . , N + 1 as an exponential of a the time-inhomogeneous L´evy process given by equation (2.2.84). This approach, however, leads to negative Libor rates (which is not necessarily inconceivable). On the other hand, it simplifies the form of the compensator of the jump measure µL under each forward measure (see equation (2.2.94)), allowing us to use the theory of Esscher transformation. This fact, in particular, makes the calibration procedure, which will be described in the next chapters, much easier allowing us to avoid a few tedious numerical calculations. Alternatively, we postulate the following dynamics of the domestic forward process under the domestic forward measure PTj
dL(t, Tj−1) with the initial condition
FB(0, Tj−1, Tj) = B(0, Tj−1)
B(0, Tj) , j = 1, . . . , N + 1.
Consequently, the Libor rate L(t, Tj−1) satisfies dL(t, Tj−1)
L(t−, Tj−1) = 1 + δL(t−, Tj−1) δL(t−, Tj−1)
λ(t, Tj−1)>ctdWtTj
+
Exactly as in the previous case dPTj−1
Thus, for j = 1, . . . , N we can establish the following connection between the domestic forward measures PTj and PT∗
Hence, the density process dPTj
dPT∗ admits the following representation:
dPTj
where CtT∗ is a deterministic function given by
One can easily check that
CtT∗ = exp