Interest Rates and The Credit Crunch

New Formulas and Market Models
Fabio Mercurio
QFR, Bloomberg∗
First version: 12 November 2008
This version: 5 February 2009

We start by describing the major changes that occurred in the quotes of market
rates after the 2007 subprime mortgage crisis. We comment on their lost analogies
and consistencies, and hint on a possible, simple way to formally reconcile them. We
then show how to price interest rate swaps under the new market practice of using
different curves for generating future LIBOR rates and for discounting cash flows.
Straightforward modifications of the market formulas for caps and swaptions will also
be derived.
Finally, we will introduce a new LIBOR market model, which will be based on
modeling the joint evolution of FRA rates and forward rates belonging to the discount
curve. We will start by analyzing the basic lognormal case and then add stochastic
volatility. The dynamics of FRA rates under different measures will be obtained and
closed form formulas for caplets and swaptions derived in the lognormal and Heston
(1993) cases.



Before the credit crunch of 2007, the interest rates quoted in the market showed typical
consistencies that we learned on books. We knew that a floating rate bond, where rates are
set at the beginning of their application period and paid at the end, is always worth par
at inception, irrespectively of the length of the underlying rate (as soon as the payment
schedule is re-adjusted accordingly). For instance, Hull (2002) recites: “The floating-rate
bond underlying the swap pays LIBOR. As a result, the value of this bond equals the swap

Stimulating discussions with Peter Carr, Bjorn Flesaker and Antonio Castagna are gratefully acknowledged. The author also thanks Marco Bianchetti and Massimo Morini for their helpful comments and
Paola Mosconi and Sabrina Dvorski for proofreading the article’s first draft. Needless to say, all errors are
the author’s responsibility.


principal.” We also knew that a forward rate agreement (FRA) could be replicated by going
long a deposit and selling short another with maturities equal to the FRA’s maturity and
reset time.
These consistencies between rates allowed the construction of a well-defined zero-coupon
curve, typically using bootstrapping techniques in conjunction with interpolation methods.1
Differences between similar rates were present in the market, but generally regarded as
negligible. For instance, deposit rates and OIS (EONIA) rates for the same maturity would
chase each other, but keeping a safety distance (the basis) of a few basis points. Similarly,
swap rates with the same maturity, but based on different lengths for the underlying
floating rates, would be quoted at a non-zero (but again negligible) spread.
Then, August 2007 arrived, and our convictions became to weaver. The liquidity crisis
widened the basis, so that market rates that were consistent with each other suddenly
revealed a degree of incompatibility that worsened as time passed by. For instance, the
forward rates implied by two consecutive deposits became different than the quoted FRA
rates or the forward rates implied by OIS (EONIA) quotes. Remarkably, this divergence
in values does not create arbitrage opportunities when credit or liquidity issues are taken
into account. As an example, a swap rate based on semiannual payments of the six-month
LIBOR rate can be different (and higher) than the same-maturity swap rate based on
quarterly payments of the three-month LIBOR rate.
These stylized facts suggest that the consistent construction of a yield curve is possible
only thanks to credit and liquidity theories justifying the simultaneous existence of different
values for same-tenor market rates. Morini (2008) is, to our knowledge, the first to design
a theoretical framework that motivates the divergence in value of such rates. To this end,
he introduces a stochastic default probability and, assuming no liquidity risk and that the
risk in the FRA contract exceeds that in the LIBOR rates, obtains patterns similar to the
market’s.2 However, while waiting for a combined credit-liquidity theory to be produced
and become effective, practitioners seem to agree on an empirical approach, which is based
on the construction of as many curves as possible rate lengths (e.g. 1m, 3m, 6m, 1y).
Future cash flows are thus generated through the curves associated to the underlying rates
and then discounted by another curve, which we term “discount curve”.
Assuming different curves for different rate lengths, however, immediately invalidates
the classic pricing approaches, which were built on the cornerstone of a unique, and fully
consistent, zero-coupon curve, used both in the generation of future cash flows and in the
calculation of their present value. This paper shows how to generalize the main (interest
rate) market models so as to account for the new market practice of using multiple curves
for each single currency.
The valuation of interest rate derivatives under different curves for generating future
rates and for discounting received little attention in the (non-credit related) financial lit1

The bootstrapping aimed at inferring the discount factors (zero-coupon bond prices) for the market
maturities (pillars). Interpolation methods were needed to obtain interest rate values between two market
pillars or outside the quoted interval.
We also hint at a possible solution in Section 2.2. Compared to Morini, we consider simplified assumptions on defaults, but allow the interbank counterparty to change over time.

erature, and mainly concerning the valuation of cross currency swaps, see Fruchard et
al. (1995), Boenkost and Schmidt (2005) and Kijima et al. (2008). To our knowledge,
Bianchetti (2008) is the first to apply the methodology to the single currency case. In this
article, we start from the approach proposed by Kijima et al. (2008), and show how to
extend accordingly the (single currency) LIBOR market model (LMM).
Our extended version of the LMM is based on the joint evolution of FRA rates, namely
of the fixed rates that give zero value to the related forward rate agreements.3 In the
single-curve case, an FRA rate can be defined by the expectation of the corresponding
LIBOR rate under a given forward measure, see e.g. Brigo and Mercurio (2006). In our
multi-curve setting, an analogous definition applies, but with the complication that the
LIBOR rate and the forward measure belong, in general, to different curves. FRA rates
thus become different objects than the LIBOR rates they originate from, and as such
can be modeled with their own dynamics. In fact, FRA rates are martingales under the
associated forward measure for the discount curve, but modeling their joint evolution is
not equivalent to defining their instantaneous covariation structure. In this article, we will
start by considering the basic example of lognormal dynamics and then introduce general
stochastic volatility processes. The dynamics of FRA rates under non-canonical measures
will be shown to be similar to those in the classic LMM. The main difference is given by
the drift rates that depend on the relevant forward rates for the discount curve, rather
then the other FRA rates in the considered family.
A last remark is in order. Also when we price interest rate derivatives under credit
risk we eventually deal with two curves, one for generating cash flows and the other for
discounting, see e.g. the LMM of Sch¨onbucher (2000). However, in this article we do
not want to model the yield curve of a given risky issuer or counterparty. We rather
acknowledge that distinct rates in the market account for different credit or liquidity effects,
and we start from this stylized fact to build a new LMM consistent with it.
The article is organized as follows. Section 2 briefly describes the changes in the main
interest rate quotes occurred after August 2007, proposing a simple formal explanation
for their differences. It also describes the market practice of building different curves and
motivates the approach we follow in the article. Section 3 introduces the main definitions
and notations. Section 4 shows how to value interest rate swaps when future LIBOR rates
are generated with a corresponding yield curve but discounted with another. Section 5
extends the market Black formulas for caplets and swaptions to the double-curve case.
Section 6 introduces the extended lognormal LIBOR market model and derives the FRA
and forward rates dynamics under different measures and the pricing formulas for caplets
and swaptions. Section 7 introduces stochastic volatility and derives the dynamics of rates
and volatilities under generic forward and swap measures. Hints on the derivation of pricing
formulas for caps and swaptions are then provided in the specific case of the Wu and Zhang
(2006) model. Section 8 concludes the article.
These forward rate agreements are actually swaplets, in that, contrary to market FRAs, they pay at
the end of the application period.

2005 to November 12. All these rates. from November 14th. but different floating legs (in terms of payment frequency and length of the paid rate). In Figure 2 we compare the “last” values of two two-year swap rates. 4. whereas the value of the corresponding FRA rate was 1. for instance.357%.4 Historical values of some relevant rates are shown in Figures 1 and 2. 2008 were. tend to be rather close to the corresponding FRA rates. let us denote the FRA rate and the forward rate implied by the two deposits with maturity T1 and T2 by FX and FD .4 2 Credit-crunch interest-rate quotes An immediate consequence of the 2007 credit crunch was the divergence of rates that until then closely chased each other. T1 ) 4 Futures rates are less straightforward to compare because of their fixed IMM maturities and their implicit convexity correction. We can see that the basis was well below ten bp until August 2007. which we learnt to be equal to the forward rate implied by two related deposits. these values do not necessarily lead to arbitrage opportunities.5% lower. deposit and OIS rates with the same maturity. T2 ]): a) Buy (1 + τ1. for instance. In fact. but since then started moving erratically around different levels. . FRA rates.345%. which were so closely interconnected. Again. however.2 is the year fraction for (T1 . quoted at 2. but diverged heavily thereafter. each one incorporating its own liquidity or credit premium. the second paying semiannually the six-month LIBOR rate. Assuming. Rates implied by other market quotes are. 2008. Rates related to the same time interval are. Their values. 2.2 FD )D(0. Once again.85%. paying (1 + τ1. and assume that FD > FX . In Figure 3 we compare the “last” values of 3x6 EONIA forward rates and 3x6 FRA rates. we can notice the change in behavior occurred in August 2007. the first paying quarterly the three-month LIBOR rate. T2 ) = D(0. 2008. not displaying the large discrepancies observed with other rates. for simplicity. 30/360 as day-count convention (the actual one for the EUR LIBOR rate is ACT/360). either because related to the same time interval or because implied by other market quotes. from November 14th. the implied three-month forward rate in three months is 4. respectively. One may then be tempted to implement the following strategy (τ1. suddenly became different objects. In Figure 1 we compare the “last” values of one-month EONIA rates and one-month deposit rates. 2008. from November 14th.1 Divergence between FRA rates and forward rates implied by deposits The closing values of the three-month and six-month deposits on November 12. 2005 to November 12. Another example is given by swap rates with the same maturity. respectively. 2005 to November 12. these rates have been rather aligned until August 2007.286% and 4.2 FD ) bonds with maturity T2 . Surprisingly enough.

since a zero investment today produces a (stochastic but) positive gain at time T1 or. remaining with 1 + τ1. T2 ) which is negative if rates are assumed to be positive.2 L(T1 . T2 ) 1 + τ1.2 FX −1=− . T ) denotes the time-0 bond price for maturity T . T2 ) in cash at T1 . T2 ) 1 + τ1.2 FD 1 + τ1.5 Figure 1: Euro 1m EONIA rates vs 1m deposit rates. equivalently.2 (FD − FX ) − = >0 1 + τ1.2 L(T1 . from 14 Nov 2005 to 12 Nov 2008. dollars.2 FX τ1. Source: Bloomberg. paying out at time T1 τ1. At time T1 . T2 ) 1 + τ1.2 FD bonds with maturity T2 .2 (L(T1 . a deterministic positive gain at T2 (with no intermediate net cash . b) plus c) yield τ1. where D(0. T2 ) − FX ) 1 + τ1.2 (FD − FX ) received at 2 . T2 ) − FX ) 1 + τ1. To pay this residual debt. T2 ) is the LIBOR rate set at T1 for maturity T2 . T2 ) where L(T1 . The value of this strategy at the current time is zero. T1 ) dollars.2 L(T1 .2 L(T1 . This is clearly an arbitrage. b) Sell 1 bond with maturity T1 .2 L(T1 . which is equivalent to τ1. 1 + τ1. we sell the 1 + τ1.2 L(T1 . receiving D(0. c) Enter a (payer) FRA.2 (L(T1 .

His interbank counterparty is.2 Explaining the difference in value of similar rates The difference in value between formerly equivalent rates can be explained by means of a simple credit model. If either events occur.5 Therefore.6 To this end. there are two issues that. we can end up with a loss at final time T2 that may outvalue the positive gain τ1. we can conclude that the strategy above does not necessarily constitute an arbitrage opportunity. . however. ii) Possibility of liquidity crunch at times 0 or T1 . Source: Bloomberg. he considers bilateral default risk. kept the same.6 Figure 2: Euro 2y swap rates (3m vs 6m). 6 Morini (2008) develops a similar approach with stochastic probability of default. However. The forward rates FD and FX are in fact “allowed” to diverge. default still plays against us. 2. and his definition of FRA contract is different than that used by the market.2 (FD − FX ). which is based on assuming that the generic interbank counterparty is subject to default risk. can not be neglected any more (we assume that the FRA is default-free): i) Possibility of default before T2 of the counterparty we lent money to. in the current market environment. from 14 Nov 2005 to 12 Nov 2008. let us denote by τt the default time of the generic 5 Even assuming we can sell back at T1 the T2 -bonds to the counterparty we initially lent money to. and their difference can be seen as representative of the market estimate of future credit and liquidity issues. In addition to ours. flows).

where E denotes expectation under the risk-neutral measure.7 Figure 3: 3x6 EONIA forward rates vs 3x6 FRA rates. T )E 1{τt >T } |Ft . T ) = E e− t r(u) du R+(1−R)1{τt >T } |Ft = RP (t. Source: Bloomberg. r the default-free instantaneous interest rate. which is the simple interest earned by the deposit D(T1 . P (t. T )+(1−R)P (t.7 Setting   Q(t.2 D(T1 .2 P (T1 . from 14 Nov 2005 to 12 Nov 2008. T2 ) = τ1. interbank counterparty at time t. T ) := E 1{τt >T } |Ft . the value at time t of a deposit starting at that time and with maturity T is h RT  i   D(t. T2 ). T2 ) R + (1 − R)Q(T1 . Assuming independence between default and interest rates and denoting by R the (assumed constant) recovery rate. T2 ) τ1. L(T1 . is given by     1 1 1 1 1 −1 = −1 . . T2 ) 7 We also refer to the next section for all definitions and notations. where the subscript t indicates that the random variable τt can be different at different times. the LIBOR rate L(T1 . T2 ). T ) the price of a default-free zero coupon bond at time t for maturity T and Ft is the information available in the market at time t.

T2 ) R + (1 − R)E Q(T1 .8 Assuming that the above FRA has no counterparty risk. P (0. T2 )   R  T1 1 + τ1. i. T2 ) will be larger than the FRA rate FX if R + (1 − R)Q(0. > R + (1 − R)Q(0. T1 ) and P (0. the FRA rate FX is larger than the forward rate implied by the default-free bonds P (0.2 FX )P (0. T2 ) Since 0 ≤ R ≤ 1. T1 ) and D(0. If the OIS (EONIA) swap curve is elected to be the risk-free curve.2 P (0. the forward rate implied by the two deposits D(0. T1 ) 1  .     1 R + (1 − R)Q(0. then (1) explains that the FRA rate FX can be (arbitrarily) higher than the corresponding forward OIS rate if the default risk implicit in the LIBOR rate is taken into account. T2 ) (1) Therefore. T1 ) − (1 + τ1. T2 ) which yields the value of the FRA rate FX :   1 P (0. refer to different default times τ0 and τT1 . T2 ) This happens.2 L(T1 . T1 ) 1   −1 . Ti ). T2 ) R + (1 − R)E Q(T1 . .2 FX )P (T1 . T1 ) 1 D(0.e. T1 ) −1 . T2 ) R + (1 − R)E Q(T1 .2 (L(T1 . 0 < Q(T1 . they can not be regarded as completely unrelated to each other. when R < 1 and the market expectation for the future credit premium from T1 to T2 (inversely proportional to Q(T1 . T2 ) and Q(0. T2 )) is low compared to the value implied by the spot quantities Q(0. which is reasonable since the credit risk in an overnight rate is deemed to be negligible even in this new market situation. i = 1.2   P (0. 2. its time-0 value can be written as  R  T − 0 1 r(u) du τ1. FX = τ1. T2 ). T2 ) < 1 so that FX > 1 τ1. T2 ). T1 ) FD = −1 = −1 τ1. since they both depend on the credit worthiness of the generic interbank counterparty from T1 to T2 . T2 ) τ1.2 L(T1 . T2 ) h R T1 i = E e− 0 r(u) du 1 − (1 + τ1.8 8 Even though the quantities Q(T1 . then   0 < R + (1 − R)E Q(T1 . T1 ) P (0. for instance.2 R + (1 − R)Q(0. T2 )(R + (1 − R)Q(T1 .2 FX − 0 r(u) du 1− =E e 1 + τ1. T2 ))   = P (0. T2 ) < 1. T1 ) and Q(0. T2 ) P (0. Similarly. T2 ). T2 ) − FX ) 0=E e 1 + τ1.2 D(0.

9 Further degrees of freedom to be calibrated to market quotes can be added by also modeling liquidity risk. the interbank one. but because the counterparty is generic and a new default time τt is generated at each time t to assess the credit premium in the LIBOR rate at that time. This task. in fact. but distinct. given that practitioners build different curves for different tenors. by Cetin et al. since the default time does not refer to a single credit entity. The forward (or ”growth”) curve associated to a given rate tenor can be constructed with standard bootstrapping techniques. however. or any version obtained by mixing “inhomogeneous rates”. one for each possible rate length considered. (2006) and Acerbi and Scandolo (2007). was straightforward to accomplish thanks to the existence of a unique. we will model forward rates with a given tenor in conjunction with those implied by the discount curve. it is still not clear how to account for these new market features and practice. 2. before August 2007. To this end. This results in the construction of different zero-coupon curves. but it is representative of a generic sector. does not change over time because the credit worthiness of the reference entity evolves stochastically. ii) Modeling the joint. As far as derivatives pricing is concerned. The random variable τt . in fact. However. When dealing with multiple curves. that the standard theories on credit risk do not immediately apply here. however.9 A thorough and sensible treatment of liquidity effects. well defined yield curve. To this end. . become compatible with each other as soon as credit and liquidity risks are taken into account. When pricing interest rate derivatives with a given model. devising a sensible model for the evolution of default times may not be so obvious. instead of explicitly modeling credit and liquidity effects. The main difference with the methodology fol9 Liquidity effects are modeled. is then elected to act as the discount curve. practitioners seem to deal with the above discrepancies by segmenting market rates. among others. one may identify two possible solutions: i) Modeling default-free rates in conjunction with default times τt and/or liquidity effects. The former choice is consistent with the above procedure to justify the simultaneous existence of formerly equivalent rates. is however beyond the scope of this work. This will be achieved in the spirit of Kijima et al (2008). Notice. In this article. In fact. Such rates. One of this curves. therefore. evolution of rates that applies to the same interval.3 Using multiple curves The analysis just performed is meant to provide a simple theoretical justification for the current divergence of market rates that refer to the same time interval. labeling them differently according to their application period. not only the calibration to market rates but also the modeling of their evolution becomes a non-trivial task. it is quite reasonable to introduce an interest rate model where such curves are modeled jointly but distinctly. However. the usual first step is the model calibration to the term structure of market rates. we prefer to follow the latter approach and apply a logic similar to that used in the yield curves construction.

S) := τx (T. we have selected N different interest-rate lengths δ1 .}. 11 A detailed description of a possible methodology for constructing forward and discount curves is outlined in Ametrano and Bianchetti (2008). . whose best proxy is the OIS swap curve. T ) the associated discount factor (equivalently. zero-coupon bond price) at time t for maturity T . we will hint at a possible bootstrap methodology. In general. . We also assume we are given a curve D for discounting future cash flows. 12 In the next section. instead. must be discounted at LIBOR. For instance. T ) the curve-D discount factor at time t for maturity T .10 lowed in the pre-credit-crunch situation is that now only the market quotes corresponding to the given tenor are employed in the stripping procedure. . .12 We denote by Pi (t. in absence of counterparty risk or in case of collateralized derivatives. . In such a case. T1i . the discount curve may be bootstrapped (and extrapolated) from the quoted deposit rates. S) 10 Notice that OIS rates carry the credit risk of an overnight rate. Forward rates can be defined for each given curve. The extension to a more general case involves a heavier notation and here neglected for simplicity. which is the rate reflecting the credit risk of the interbank sector. . for the same currency. involves plenty of technicalities and subjective choices.10 For a contract signed with a generic interbank counterparty without collateral. bootstrapping multiple curves. The discount curve. the discount curve should reflect the fact that future cash flows are at risk and. we will assume that future cash flows are all discounted with the same discount curve. the futures (or 3m FRAs) for the main maturities and the liquid swaps (vs 3m). the three-month curve can be constructed by bootstrapping zero-coupon rates from the market quotes of the three-month deposit. . . We will consider the time structures {T0i . . (2) Fx (t. which includes the payment times of a swap’s fixed leg. can be selected in several different ways. For instance.}. . In general. therefore. .11 In the following. N. T1S . the (simply-compounded) forward rate prevailing at time t and applied to the future time interval [T. . D}. T. We denote by PD (t. obtained by suitably interpolating and extrapolating OIS swap quotes. 2. 3 Basic definitions and notation Let us assume that. The curve associated to length δi will be shortly referred to as curve i. as such. where the superscript i denotes the curve it belongs to. in a single currency economy. S] is defined by   Px (t. for each curve x ∈ {1. and {T0S . Precisely. T ) 1 −1 . the discount curve can be selected as the yield curve associated the counterparty in question. S) Px (t. δN and constructed the corresponding yield curves. . . which may be regarded as negligible in most situations. depending on the contract to price. . it can be deemed to be the classic risk-neutral curve.

TM T is the discretely-rebalanced bank account BD : PD (t. if i denotes the three-month curve. Tki ) i i where τkx is the year fraction for the interval [Tk−1 . S] under the convention of curve x. where again the indices x and z identify. . j=c+1 S . . respectively. N. . Tk−1 i i x Fk (t) := Fx (t.d CD (t) = d X τjS PD (t. S) is the year fraction for the interval [T. i i j=0 PD (Tj−1 . M. we show how to value linear interest rate derivatives under our assumption of distinct forward and discount curves. for curves i = 1. . and the superscript z defines the measure in question.13 i < Tki and the curve x ∈ {1.14 and an interest rate swap where the floating leg pays at each 13 In practice. we will make use of Given the times t ≤ Tk−1 the following short-hand notation:   i ) 1 Px (t. Tk ) = x −1 (3) τk Px (t. TjS ). As in Kijima et al (2008). whose numeraire is the annuity c. S S S • Qc. N . . . Tki ). The information available in the market at each time t will be described by the filtration Ft . . . . . More precisely. we denote by: • QTD the T -forward measure. D}. Tk−1 . 14 For instance. the pricing measures we will consider are those associated to the discount curve D. then the times Tki must be three-month spaced. . Tmi ) . T ). . T ] may be totally arbitrary. where τjS := τD (Tj−1 The expectation under the generic measure Qzx will be denoted by Exz . . where the subscript x (mainly D) identifies the underlying yield curve. TjS ). . we will consider only intervals where S = T + δi . . m = 1. i } .d D the forward swap measure defined by the time structure {Tc . . . . . Tbi compatible with curve i. namely τkx := τx (Tk−1 . Tc+1 . whose numeraire • QTD the spot LIBOR measure associated to times T = {T0i . Tki ] for curve x. To this end.11 where τx (T. . . . Td }. . 4 The valuation of interest rate swaps In this section. the underlying yield curve and the measure in question. let us consider a set of times Tai . Tj ) T BD (t) = Qm i Tm−1 < t ≤ Tmi . whereas for curve D the interval [S. whose numeraire is the zero-coupon bond PD (·. . To denote these measures we will adopt the notation Qzx . .

Tki ) = EDk Fki (Tk−1 )|Ft . Therefore. (5) In the classic single curve valuation (i ≡ D). thanks to the presence of collaterals or netting clauses. FL(t. Tk−1 . b. Tki ) = τki PD (t. .e  Ti i i Lik (t) := FRA(t.12 i time Tki the LIBOR rate of curve i set at the previous time Tk−1 . Tki ) = τki PD (t. the forward rate Fki is a martingale under Ti the associated Tki -forward measure (coinciding with QDk ). . Tki ). k = a + 1. FL(t. as is well known. Tki ) Pi (Tk−1 − 1. . however. in general. curve D can be assumed to be the risk-free one (as obtained from OIS swap rates). Tk−1 . the present value of each payment in the swap’s floating leg can be simplified as follows: i i . In the situation we are dealing with. Defining the time-t FRA rate as the fixed rate to be exchanged at time Tki for the floating payment (4) so that the swap has zero value at time t. (4) i . Tki )Lik (t) = τki Pi (t. curves i and D are different. Tki )Fki (t) = Pi (t. Tki ). Tki )Lik (t). Tk−1 ) − Pi (t. Tk−1 . Tki )EDk Fki (Tk−1 )|Ft . Accordingly. but by discounting the corresponding FRA rate. of such a payoff can be obtained by taking the discounted The time-t value. 15 For most swaps. Tk−1 Ti expectation under the forward measure QDk :15  Ti i i FL(t.16 i. and the FRA rate Lik (t) is different from Fki (t). the present value of a future LIBOR rate is no longer obtained by discounting the corresponding forward rate. Ti The forward rate Fki is not a martingale under the forward measure QDk . . the time-Tk payoff of the floating leg is i i FL(Tki . In i formulas. Tk−1 . . Tk−1 i which leads to the classic result that the LIBOR rate set at time Tk−1 and paid at time i i Tk can be replicated by a long position in a zero-coupon bond expiring at time Tk−1 and a i short position in another bond with maturity Tk . This slight abuse of terminology is justified by the definition that applies when payments occur at the end of the application period (like in this case). Tki ) = τki Fki (Tk−1 )= 1 i . so that the expected value Lik (t) coincides with the current forward rate: Lik (t) = Fki (t). Tki ) = τki PD (t. 16 This FRA rate is slightly different than that defined by the market. we can write i FL(t. see Section 2.2.

. K. .d (t) = P . TjS ). . TdS . The present value of these payments is immediately obtained by discounting them with the discount curve D: d X τjS KPD (t. Tai . We get: Pb i i i k=a+1 τk PD (t. will be different in general than PD (t. . . j=c+1 S . In the particular case of a spot-starting swap. the swap rate becomes: Pb i i i i k=1 τk PD (0. . Tj ) where L1 (0) is the constant first floating payment (known at time 0). . where we remember that τjS = τD (Tj−1 Therefore. respectively. T ) D j j j=c+1 This is the forward swap rate of an interest rate swap where cash flows are generated through curve i and discounted with curve D. Tk )Lk (0) . . Tki ) = k=a+1 b X τki PD (t. (7) d S S τ P (t. . for the reasons just explained. Tki )Lik (t).13 The net present value of the swap’s floating leg is simply given by summing the values (5) of single payments: FL(t. Tbi . . by T1i . . Tbi and T1S . (8) S0. TjS ). Tk )Lk (t) i Sa. . this is not a problem. Tbi ) or Pi (t.d (0) = P d S S j=1 τj PD (0. Tk−1 . . . Tbi ). TcS ) = b X k=a+1 τki PD (t. Remark 1 As traditionally done in any bootstrapping algorithm. . . TdS . Tai ) − Pi (t. is given by IRS(t. to the fixed-rate payer. . with Tbi = TdS . since the only requirement for quoted spot-starting swaps is that their net present value must be equal to zero. . Tai . However. Tai ) − PD (t. Tbi ) = b X i FL(t. which by .b. neither leg of a spot-starting swap needs be worth par (when a fictitious exchange of notionals is introduced at maturity). As already noticed by Kijima et at. . TjS ). with payment times for the floating and fixed legs given. TcS . .c. (6) k=a+1 which. . (2008). equation (8) can be used to infer the expected rates Lik implied by the market quotes of spot-starting swaps. Tki )Lik (t) − K d X τjS PD (t.0. . TjS ) =K j=c+1 d X τjS PD (t.b. the interest rate swap value. Let us then consider the swap’s fixed leg and denote by K the fixed rate paid on the fixed leg’s dates TcS . . j=c+1 We can then calculate the corresponding forward swap rate as the fixed rate K that makes the IRS value equal to zero at time t. .

than those obtained through classic bootstrapping methods applied to swap rates 1 − PD (0. 17 It is worth mentioning that the first proof that Black-like formulas for caps and swaptions are arbitrage free is due to Jamshidian (1996). respectively. As is well known.17 To be able to adapt such formulas to our double-curve case. Tki )Ei k [Fki (Tk−1 ) − K]+ |Ft according to the chosen dynamics. Tk−1 . . S S τ P (0. TdS ) S0. to price other swaps based on curve i. K. (1997) and Miltersen et al. t ≤ Tk−1 . approach. in conjunction with any interpolation tool. in general. the choice of the discount curve D depends on the credit worthiness of the counterparty and on the possible presence of a collateral mitigating the credit risk exposure. and then calculates the time-t caplet price Ti i i Cplt(t. and more general. Again. As already noticed by Boenkost and Schmidt (2005) and by Kijima et al. we will have to reformulate accordingly the corresponding market models. these other swaps will have different values. one notices that the forward rate Fki is Ti a martingale under the Tki -forward measure Qi k for curve i. the purpose of this section is to derive pricing formulas for options on the main interest rates. (2008). 18 We will use the symbol “d” to denote differentials as opposed to d. (1997) and the lognormal swap model of Jamshidian (1997). which will result in modifications of the corresponding Black-like formulas governed by our double-curve paradigm. The bootstrapped Lik can then be used.1 Market formula for caplets We first consider the case of a caplet paying out at time Tki i τki [Fki (Tk−1 ) − K]+ . the classic choice of a driftless geometric Brownian motion18 i dFki (t) = σk Fki (t) dZk (t).14 definition have zero value. T ) D j j=1 j However. this is perfectly reasonable since we are here using an alternative.d (0) = Pd . (9) To price such payoff in the basic single-curve case. which instead denotes the index of the final date in the swap’s fixed leg. 5 The pricing of caplets and swaptions Similarly to what we just did for interest rate swaps. Tki ) = τki Pi (t. For instance. the formal justifications for the use of Black-like formulas for caps and swaptions come. from the lognormal LMM of Brace et al. 5.

Fk (t). Tk ) Bl K. This is the approach proposed by Bianchetti (2008). Tk−1 . we can value the last expectation analytically and . Tki )EDk [Lik (Tk−1 ) − K]+ |Ft . replace the payoff (9) with i ) − K]+ τki [Lik (Tk−1 (11) i i and view the caplet as a call option no more on Fki (Tk−1 ) but on Lik (Tk−1 ). Since  Ti i )|Ft . the trick was to replace the LIBOR rate entering the caplet payoff with the equivalent forward rate. namely to introduce an underlying asset whose dynamics is easier to model. a martingale under the measure QDk . Tki ) = τki PD (t. Lik (t) = EDk Fki (Tk−1 i at the reset time Tk−1 the two rates Fki and Lik coincides: i i ). Our idea is to follow a conceptually similar approach as in the classic LMM. A possible way to value it is Ti to model the dynamics of Fki under its own measure Qi k and then to model the RadonTi Ti Ti Ti Nikodym derivative dQi k /dQDk that defines the measure change from Qi k to QDk . instead. Tk−1 . This leads to: Ti i i Cplt(t. In our double-curve setting. K. Here. in general. the problem with this new expectation is that the Ti forward rate Fki is not. Cplt(t.15 Ti where σk is a constant and Zk is a Qi k -Brownian motion. we take a different route. the caplet valuation requires more attention. σk Tk−1 where ln(F/K) + v 2 /2 Bl(K. since Ti the pricing measure is now the forward measure QDk for curve D. we make a step forward. F. v  and Φ denotes the standard normal distribution function. and replace the forward rate with its conditional expected value (the FRA rate). by definition. Tk ) = τk Pi (t. There. who uses a foreign-currency analogy and derives a quanto-like correction for the drift of Fki . Tk−1 . Tki ) = τki PD (t. ) = Fki (Tk−1 Lik (Tk−1 We can. leads to Black’s pricing formula: q  i i i i i i −t (10) Cplt(t. therefore. Here. Tki )EDk [Fki (Tk−1 As already explained in the IRS case. K. K. the caplet price at time t becomes Ti i i ) − K]+ |Ft . a martingale under QDk . The purpose is the same as before. v) = F Φ v    ln(F/K) − v 2 /2 − KΦ . Ti (12) The FRA rate Lik (t) is. If we smartly choose the dynamics of such a rate. since the latter has “better” dynamics (a martingale) under the reference pricing measure. In fact.

d D . Tk ) Bl K. In fact.b. which here belongs to curve D. also in our multi-curve paradigm. . . Tbi . (13) j=c+1 where. Its payoff at time Ta = Tc is therefore d  i + X i Sa. we get: i S PS(t.d = |Ft τj PD (t. .16 obtain a closed-form formula for the caplet price. 5. the caplet price is again given by Black’s formula with an implied volatility vk . . whose associated c. The differences with respect to the classic formula (10) are given by the underlying rate. . . K. see (7). .b.d (Tai ) − K j=c+1 τj PD (Tc . A payer swaption gives the right to enter at time Tai = TcS an IRS with payment times for the i S floating and fixed legs given. Tk ) = τk PD (t. . .d CD (t) = d X τjS PD (t.c. Tj ) S S Qc. which here is the FRA rate Lik . leads to the following pricing formula: q  i i i i i i −t . pricing a swaption is equivalent to pricing an option on the underlying swap rate. Cplt(t. Tbi and Tc+1 . the obvious choice of a driftless geometric Brownian motion dLik (t) = vk Lik (t) dZk (t). . . S S τ P (t.d τjS PD (t. .d numeraire is the annuity CD (t).c. vk Tk−1 Therefore. by Ta+1 . .d S) C (T c D j=c+1 = d X c. TdS . under lognormal dynamics for the rate Lik . with Tbi = TdS i S and where the fixed rate is K. Pb i Sa.b.d (Ta ) − K τjS PD (TcS . respectively. TjS ) j=c+1 the payoff (13) is conveniently priced under the swap measure Qc. K. TdS ) + Pd  i  d S S S X Sa. .b.d (t) i i i k=a+1 τk PD (t. and by the discount factor. Tj ) E D c.d (Tai ) − K |Ft j=c+1 so that. . Tc+1 . T ) D j j j=c+1 = Pd Setting c. Tk )Lk (t) .2 Market formula for swaptions The other plain-vanilla option in the interest rate market is the European swaption. i t ≤ Tk−1 Ti where vk is a constant and Zk is now a QDk -Brownian motion. Lk (t). .c. TjS ) E QD (14)  i + Sa. Ta+1 .c. Tk−1 . For instance. TjS ). .

d = i i i k=a+1 τk PD (t. . In fact.d (t).d (t) has a more general definition. . .b.b. .d p  Tai − t .d Assuming that the swap rate Sa. .b.c. leading to the generalized Black formula: i S PS(t. . . Using measure change techniques. To extend the LMM to the multi-curve case.d is a constant and Za. . we are now ready to extend the basic LMMs.c. under Qc. j=c+1 Therefore. TM } the times in question. one then jointly Ti i models rates Fk . the double-curve swaption price is still given by a Black-like formula. Ta+1 .b. TdS ) = d X i τjS PD (t.b. TjS ) Bl K.c.b.c. Tj ) = FL(t.d D -Brownian motion. one finally derives pricing formulas for the main calibration instruments (caps and swaptions) either in closed form or through efficient approximations.d (t) is a martingale under the c.d (t).c. t ≤ Tai where νa. Tbi ) c. We start by considering the fundamental case of lognormal dynamics.c.c. which depends on LIBOR .b. Tai . Sa. .17 i As in the single-curve case.d Sa. Tk )Lk (t) Pd S S j=c+1 τj PD (t. . .b.b. with the only differences with respect to the basic case that discounting is done through curve D i and that the swap rate Sa. the expectation in (14) can be explicitly calculated as in the caplet case. The previous section suggests that the FRA rates Lik are convenient rates to model as soon as we have to price a payoff. like that of a caplet.b. Denoting by T = {T0i . . .d D . .c.d is a Qc. .c.d of the swap) divided by the numeraire CD (t): Pb i (t) Sa.d (t) dZa. M . . K.b.b. by (6).d i swap measure QD . under the forward measure Qi M or under the spot LIBOR measure QTi . (15) i evolves. . Tc+1 .c. 6 The double-curve lognormal LMM In the classic (single-curve) LMM.c. Tbi . νa.b.c. and then introduce stochastic volatility in a rather general fashion. according to a driftless geometric Brownian motion: i i dSa. After having derived market formulas for caps and swaptions under distinct discount and forward curves. typically some “terminal” forward measure or the spot LIBOR measure corresponding to the set of times defining the family i of forward rates. the forward swap rate Sa.d CD (t) . . one models the joint evolution of a set of consecutive forward LIBOR rates under a common pricing measure.d (t) is equal to a tradable asset (the floating leg c. . Sa. k = 1. we first need to identify the rates we need to model. .d (t) = νa.c.

. 6. i t ≤ Tk−1 (18) where the instantaneous volatility σk (t) is deterministic and Zk is the k-th component of an Ti M -dimensional QDk -Brownian motion Z with instantaneous correlation matrix (ρk.c. (1997) and Miltersen et al. namely dZk (t) dZj (t) = ρk.. (1997) by assuming that forward LIBOR rates have a lognormal-type diffusion coefficient..d (t) = Pd = ωk (t)Lik (t). unless we unrealistically assume a deterministic discount curve. Notice. Tj ) . Moreover. i Let us consider a set of times T = {0 < T0i .M . . Moreover. In fact.. as is evident from equation (16).j dt. Tk )Lk (t) i Sa. under its canonical forward measure Ti QDk . . T ) j j=c+1 j D k=a+1 where the weights ωk are defined by ωk (t) := Pd τki PD (t.1 The model dynamics The LMM was introduced in the financial literature by Brace et al. also the consistency with the standard single-curve case. we extend their approach to the case where the curve used for discounting is different than that used to generate the relevant future rates.18 rates belonging to the same curve i. where the forward LIBOR rates Fki (t) and the FRA rates Lik (t) coincide by definition. future swap rates also depend on future discount factors which. (17) i (t) as a linear combination of FRA rates Lik (t) Characterizing the forward swap rate Sa. so that also path-dependent payoffs on LIBOR rates will depend on the dynamics of the discount curve.j=1.19 Here.d gives another argument supporting the modeling of FRA rates as fundamental bricks to generate sensible future payoffs in the pricing of interest rate derivatives.c. there is a major difference with respect to the single-curve case. will evolve stochastically over time.b. we stick to the case where these rates belong to the same curve i.b. namely that forward rates belonging to the discount curve need to be modeled too. We assume that each rate Lk (t) evolves. which we assume to be compatible i with curve i.j )k. 19 This implies that each forward LIBOR rate evolves according to a geometric Brownian motion under its associated forward measure. . we notice we can write Pb b i i i X k=a+1 τk PD (t. . as a driftless geometric Brownian motion: dLik (t) = σk (t)Lik (t) dZk (t). Tki ) S S j=c+1 τj PD (t. However. in case of a swap-rate dependent payoff. For simplicity. (16) S S τ P (t. we will show below that the dynamics of FRA rates under typical pricing measures depend on forward rates of curve D. TM }. .

. The analysis that follows can be equivalently applied to the new dynamics of rates FkD .h dt i.h dt dZk (t) dZhD (t) = ρi. for instance. we assume that the dynamics of each rate FhD under the associated forward Ti measure QDh is given by: dFhD (t) = σhD (t)FhD (t) dZhD (t).D = (ρi. .D ρ 0 R := ρi. Tki ) τk i τkD = τD (Tk−1 . one can assume that each spread Xk (t) := |Lik (t) − FkD (t)| evolves under Ti the corresponding forward measure QDk . the dynamics (19) of forward rates FkD must then be replaced with dFkD (t) = dLik (t) ± dXk (t). In such a case.g. ρD.D = (ρD..M must be chosen so as to ensure that the global matrix   ρi.D Clearly. when one curve is above the other and there are sound financial reasons why the spread should be preserved positive in the future.. it may be more realistic to resort to an alternative approach and model either curve i or D jointly with the spread between them.h=1. Tki ) To this end.. Remark 2 In some situations. This happens. correlations ρ = (ρk.20 20 The calculations are essentially the same.D ρD. according to some dXk (t) = σkX (t.D k.D k... see e.j=1.. whose solution is positively distributed. we also need to model the evolution of rates   i ) 1 PD (t..h )k. Xk (t)) dZkX (t)..h=1. Sticking to (18). Tk ) = D PD (t.19 In a double-curve setting.D is positive (semi)definite. i t ≤ Th−1 (19) where the instantaneous volatility σhD (t) is deterministic and ZhD is the h-th component of Ti an M -dimensional QDh -Brownian motion Z D whose correlations are dZkD (t) dZhD (t) = ρD.j )k.M and ρ k.. where the sign ± depends on the relative position of curves i and D.. Their length depends on the chosen volatility function σkX .. Tk−1 . Tk−1 D i i −1 Fk (t) = FD (t.D k. Kijima (2008) or Sch¨ onbucher (2000).h )k.M .

QDj ) dhLik . the situation is different since rates Lik and FhD belong. and to calculate the instantaneous covariations in the drift term. ln 1 + τhD FhD it =− = dt dt h=j+1 = k X h=j+1 τhD dhLik . The initial assumptions on the joint dynamics of forward rates are therefore sufficient to determine such a drift term. in general. we apply the change-of-numeraire formula relating the drifts of a given process under two measures with known numeraires.2 Dynamics under a general forward measure Ti To derive the dynamics of the FRA rate Lik (t) under the forward measure QDj we start Ti Ti from (18) and perform a change of measure from QDk to QDj . the drift term of Lik under QDj depends on the instantaneous covariations between forward rates Fki and Fhi . The log of the ratio of the two numeraires can be written as #! " k Y (1 + τhD FhD (t)) ln(PD (t. Tki )/PD (t. whose associated numeraires are the curve-D zero-coupon bonds with maturities Tki and Tji . FhD it . ln(PD (·. Tji ))it dhLik . h = j + 1. . dt 1 + τhD FhD (t) Ti In the standard LMM. respectively. Tji )) = ln 1/ h=j+1 =− k X  ln 1 + τhD FhD (t) h=j+1 from which we get: Ti Drift(Lik . Let us first consider the case j < k. D D 1 + τ F (t) h h h=j+1 The derivation of the drift rate in the case j > k is perfectly analogous. Tki )/PD (·.h τh σh (t)Fh (t) . Tji ))it =− .D k. to different curves. Y it denotes the instantaneous covariation between processes X and Y at time t. dt where hX. . Under (19).20 6. Here. To this end. k. we also need the dynamics of rates FhD . . The drift of Ti Lik (t) under QDj is then equal to Ti Drift(Lik . we thus obtain Ti Drift(Lik . however. ln(PD (·. . . Tki )/PD (·. QDj )  k X dhLik . QDj ) = σk (t)Lik (t) k D D D X ρi. see for instance Brigo and Mercurio (2006).

k. t ≤ Tk−1 : D. j = k and j > k are. we can also see that the FRA rate dynamics reduce to those of the corresponding forward rates since  i.h → ρk.h    τ D → τ i h h i≡D⇒ D  σh (t) → σh (t)    D Fh (t) → Fhi (t) for each h. we can easily prove that the SDEs (20) for the FRA rates all admit a unique strong solution if the coefficients σhD are bounded. see Brigo and Mercurio (2006).D (t) (t)F σ τ  h h h k.  " k # D D D  X ρi. .D with correlation matrix R.D D D D X ρ τ σ (t)F (t)  h k. we have already noticed that the FRA rates Lik coincide with the corresponding Fki . D under The joint evolution of all FRA rates Li1 .D ρk.D (t) " #  j i. . FM a common forward measure is then summarized in the following. t ≤ Tk−1 : dFkD (t) = σkD (t)FkD (t) dZkj. . respectively.D (t) dFk (t) = σk (t)Fk (t) − D D 1 + τ F (t) h h h=k+1 (20) Ti where Zkj and Zkj.21 Ti As to forward rates FkD . When curves i and D coincide. .h h h dF D (t) = σ D (t)F D (t)  dt + dZkj.D D D D X  ρ τ σ (t)F (t)  h k. having doubled the number of rates to simulate.h j   dLi (t) = σk (t)Lik (t) dt + dZk (t)   k 1 + τhD FhD (t) h=j+1 " k # j < k. their QDj -dynamics are equivalent to those we obtain in the classic single-curve case. Remark 4 Following the same arguments used in the standard single-curve case. the computational burden of the lognormal . t ≤ Tji : D.D (t) k k  D D  k 1 + τh Fh (t) h=j+1 ( dLik (t) = σk (t)Lik (t) dZkj (t) i j = k. since these probability measures and rates are associated to the same curve D. As a further sanity check.h τhD σhD (t)FhD (t)   D D D   dt + dZkj. . The extended dynamics (20) may raise some concern on numerical issues. . .h h h  i i  dt + dZkj (t)  D D dLk (t) = σk (t)Lk (t) − 1 + τ F (t) h h i " h=k+1j # j > k. .D X ρk.D are the k-th components of M -dimensional QDj -Brownian motions Z j and Z j. In fact. Ti Proposition 3 The dynamics of Lik and FkD under the forward measure QDj in the three cases j < k. LiM and forward rates F1D .

D (t) 1 + τhD FhD (t) (21) h=β(t) d. Tki ) Bl(K. leads to the same drift rates for Lk and the corresponding Fk . . .D k. . Proposition 5 The dynamics of FRA and forward rates under the spot LIBOR measure QTD are given by: dLik (t) = σk (t)Lik (t) k D D D X ρi.D d where Z d = {Z1d .3 Dynamics under the spot LIBOR measure Another measure commonly used for modeling the joint evolution of the given rates and for pricing related derivatives is the spot LIBOR measure QTD associated to times T = i T {T0i . 6. i i P (T . . ZM } are M -dimensional QTD -Brownian motions with correlation matrix R. so that t ∈ (Tβ(t)−2 . Tβ(t)−1 ) T BD (t) = Qβ(t)−1 . We just have to replace forward rates with FRA rates and use the discount factors coming from curve D. K. . 6.h for each h. Tk−1 . immediately leads to the following. . However. For instance. m ≥ 1.22 LMM (20) is doubled with respect to that of the single-curve case. TM } .D k.D = {Z1d. thanks to the lognormality assumption.h τh σh (t)Fh (t) D D D dFk (t) = σk (t)Fk (t) dt + σkD (t)FkD (t) dZkd.D i D ρi. Tki ) = τki PD (t.D k. Tβ(t)−1 ]. assuming that D.D . . thus halvening the number of drifts to be calculated at each simulation time. We get: i Cplt(t. T ) D j−1 j j=0 i i i i where β(t) = m if Tm−2 < t ≤ Tm−1 .h τh σh (t)Fh (t) dt + σk (t)Lik (t) dZkd (t) 1 + τhD FhD (t) h=β(t) k D D D X ρD. whose numeraire is the discretely-rebalanced bank account BD i PD (t. . Application of the change-of numeraire technique. This gives a valuable advantage since it is well known that the drift calculations in a LMM are extremely time consuming. some smart selection of the correlations between rates can reduce the simulation time. vk (t)) where s Z i Tk−1 σk (u)2 du vk (t) := t As expected. since the SDEs for the homologues Lik and FkD share the same structure. this formula for caplets (and hence caps) is analogous to that obtained in the basic lognormal LMM.4 Pricing caplets in the lognormal LMM The pricing of caplets in the LMM is straightforward and follows from the same arguments of Section 5. k. . Lik (t).h = ρk. ZM } and Z d. . . . .

c. Contrary to the single curve case. (23) we equate the instantaneous quadratic variations of (22) and (23) b X  2 i va. (24) h.k=a+1 Freezing FRA and swap rates at their time-zero value.c. Tki ) S S j=c+1 τj PD (t.d (t) = Sa.b. k=a+1 ωk (t) = Pd τki PD (t.d (t) ≈ b X ωk (0)σk (t)Lik (t) dZkc.b.b. we can resort to a standard approximation technique and freeze the weights ωk at their time-zero value.b.b.c. dSa.d (t) satisfies the S.c.b.d (t)va.23 6.b.5 Pricing swaptions in the lognormal LMM An analytical approximation for the implied volatility of swaptions can be derived also in our multi-curve setting.d (t).d (t)Sa.c. the swap rate Sa. since they also depend on discount factors calculated on curve D.k t i i dZa.d (0))2 . (22) k=a+1 i Notice.d (t) = ωk (t)Lik (t).d (t) on forward rates FhD .c.d (t) b i X ∂Sa.b. in fact.d (t).d (t) can be written as a linear combination of FRA rates Lik (t).d (t) i (Sa.c.b. To this end. under the swap measure Qc.b. This leads to the approximation i Sa.c.d (t).b.d (t) dt = ωh (0)ωk (0)σh (t)σk (t)Lih (t)Lik (t)ρh.b.d D .d (t) = σk (t)Lik (t) dZkc.D.d (t): v u Pb i i u h.b. i ∂Lk (t) k=a+1 However.d (t) dZa. that by freezing the weights.c.b. Therefore we can not i write that.c. i The forward swap rate Sa.c.b.b.c. we are also freezing the dependence of Sa.c.b. i dSa. remember (16) and (17): b X i Sa. To obtain a closed equation of type i i dSa.k dt.c.k=a+1 ωh (0)ωk (0)σh (t)σk (t)Lh (0)Lk (0)ρh.c.c.c.c.d D -dynamics for the swap rate Sa.c.d (t) = Sa.d (t). we obtain this (approximated) i Qc. k=a+1 which enables us to write i dSa.b. the weights are not a function of the FRA rates only. Tj ) .d (t) ≈ b X ωk (0)Lik (t).b.E.

Tiij ) Pd j=c+1 τj PD (t. Tai ) k Y 1 i τk D D 1 + τh Fh (t) h=a+1 D = =: f (Fa+1 (t). .c.b. . .b. In this case.d i σ (t)L (t) dZ (t) + σk (t)FkD (t) dZkc. .c. i Under the swap measure Qc.d D -Brownian motions with correlation matrix R. .c.d (t) D c.b.b. The difference here is that the swaption volatility depends both on curves i and D. FbD (t)) ij Y Pd 1 S j=c+1 τj D D 1 + τh Fh (t) a=c+1 τki where the last equality defines the function f and where the subscripts of rates FhD (t) range from a + 1 to b since TdS = Tbi (namely id = b). ZM } are M -dimensional Qc.D where {Z1c. Tc+1 .c. j=c+1 p where the swaption implied volatility (multiplied by Tai ) is given by v u b u X ωh (0)ωk (0)Lih (0)Lik (0)ρh. .d (t) = b b i i X X ∂Sa.b. . approximation for the implied swaption volatility in the lognormal LMM. .d .D.E.d . which we here omit for brevity.k Z Tai σh (t)σk (t) dt. so that. Tbi . .24 This immediately leads to the following (payer) swaption price at time 0: i S PS(0. . A better approximation for lognormal LMM swaption volatilities can be derived by i }. see Brigo and Mercurio (2006). Tki ) τ i PD (t. .c. . . TjS ) Bl K. Va. Matching instantaneous quadratic variations as in (24) and freezing stochastic quantities at their time-zero value.d = t i (0))2 (Sa.d (t) then satisfies the S. Ta+1 . . . . TM S i index ij such that Tj = Tij .b. the swap rate Sa. there exists an assuming that each TjS belongs to T = {T0i . since weights ω’s belong to curve D. for each j.c.b. ZM } and {Z1c. . K. . more accurate. T i ) PD (t. .d.d (t) ∂Sa. whereas the FRA and swap rates are calculated with both curves. k k k i D ∂L (t) ∂F (t) k k k=a+1 k=a+1 c.D (t).d c.d 0 h. .d. .d (0).d D .k=a+1 (25) Again.c. we can write: PD (t. we can finally obtain another. . this formula is analogous in structure to that obtained in the classic lognormal LMM. . Sa. .c. Tai ) ωk (t) = Pd k S k S = PD (t. Va.d.D .b. Tj ) S j=c+1 τj PD (t. i dSa. TdS ) = d X  i τjS PD (0. .

at time 0. under the forward measure QDr .l τl σl (t)Fl (0)   −σ (t)   x 1 + τ D F D (0) l l=x+1 if r < x if r = x if r > x l Lix Since νx is deterministic. x = k. we first recall the dynamics of Lix . h 2 0  Ti EDr Lik (Tji )Lih (Tji ) = Lik (0)Lih (0) exp Z 0 Tji  [νk (t) + νh (t) + ρk.h σk (t)σh (t)] dt . x = k. in the drift rate µx (t). the terminal correlation between the FRA Ti rates Lik and Lih at time Tji . h. with j ≤ k − 1 < h.6 The terminal correlation between FRA rates Assume we are interested to calculate.D  x.25 6. h 0 Ti EDr 2  Lix (Tji ) = (Lix (0))2 Z Tji exp  [2νx (t) + (σx (t)) ] dt . under QDr : dLix (t) = µx (t)Lix (t) dt + σx (t)Lix (t) dZx (t) where  x D D D X  ρi.D  x.D  x. with r ≥ j:  Ti CorrDr Lik (Tji ). we freeze the forward rates FlD (t) at their time-0 value to obtain: dLix (t) = νx (t)Lix (t) dt + σx (t)Lix (t) dZx (t) where  x D D D X  ρi. The expectations (26) are thus straightforward to calculate.l τl σl (t)Fl (0)  σ (t)  x   1 + τlD FlD (0)  l=j+1  νx (t) := 0   j D D D  X ρi.l τl σl (t)Fl (t)   −σ (t)   x 1 + τ D F D (t) l l=x+1 if r < x if r = x if r > x l Then.l τl σl (t)Fl (t)  σ (t)  x   1 + τlD FlD (t)  l=j+1  µx (t) := 0   j D D D  X ρi.D  x. We get:  Z Ti  j  Tri  i i i ED Lx (Tj ) = Lx (0) exp νx (t) dt . x = k. Ti see Brigo and Mercurio (2006). follows (approximately) a geometric Brownian motion. Lih (Tji )   Ti  Ti Ti EDr Lik (Tji )Lih (Tji ) − EDr Lik (Tji ) EDr Lih (Tji ) (26) q = q i        i i i 2 2 T T T T EDr (Lik (Tji ))2 − EDr Lik (Tji ) EDr (Lih (Tji ))2 − EDr Lih (Tji ) Mimicking the derivation of the approximation formula in the single-curve lognormal LMM.

iv) closed-form approximation for terminal correlations. 2008.h σk (t)σh (t) dt − 1  r Lik (Tji ). ii) efficient explicit approximation for swaption volatilities. The most popular extensions of the LMM are based on modeling stochastic volatility as in Heston (1993) or in Hagan et al. x = h. do not enter the analytical approximation for terminal correlations. Brigo and Mercurio (2006). the drift rates µx . To properly account for the typical shapes observed in the market. which are the only terms that change when moving from singleto double-curve LMM. iii) exact (cascade) calibration of at-the-money swaption volatilities. ii) and iv) have been shown before. The advantages are in fact shared by our double-curve extension.g. In the first category fall the LMMs by 21 Points i). v) follows immediately. In fact. Strikes from 1%to 8%. (2002). see Figures 4 and 5.21 As is also well known. whereas iii) can be proved exactly as in Brigo and Mercurio (2006). 7 Introducing stochastic volatility A vast literature has been written on the single-curve LMM. this expression coincides with that we get in the single-curve case. one has to relax the constant-volatility assumption for the dynamics of the FRA rates. Source: Bloomberg. so that (26) becomes: n R Ti o exp 0 ρk. leading to an automatic calibration to (at-the-money) caplet volatilities. however.26 Figure 4: USD cap volatilities as of November 25. Lih (Tji ) = r n R Ti o n R Ti o j 2 exp 0 (σk (t)) dt − 1 exp 0 j (σh (t))2 dt − 1 j Ti CorrDr Not surprisingly. especially on its lognormal version whose advantages are well known. . see e. which thus allows for: i) caplet prices consistent with Black’s formula. the lognormal LMM has the main drawback of producing constant implied volatilities for any given maturity. k. v) deterministic future implied volatilities.

. . .j=1. Following (21). . (28) d where ak and bk are deterministic functions of time and volatility and W d = {W1d . we will show again all calculations. Vk (t)) dt + bk (t.. Henry-Labord`ere (2007) and Mercurio . as a particular case.. ZM } is again an M -dimensional QTD -Brownian motion with instantaneous correlation matrix (ρk. . we deal with general stochastic-volatility dynamics and show how to perform the relevant measure changes in our double-curve setting.M . In the second.. Here. the LMMs by Henry-Labord`ere (2007). TM curve i. all the models we mentioned above. . Lik (t))ψt (Vk (t)) dZkd (t). For example. Source: Bloomberg. . WM } T d is an M -dimensional QD -Brownian motion. .27 Figure 5: USD swaption volatilities as of November 25. 2008. Piterbarg (2005). 1 + τhD FhD (t) (27) where φk is a deterministic function of time and rate ψt is a deterministic function of time d and volatility. . and model FRA rates dynamics under the spot LIBOR measure QTD . we will follow the same procedure as in the lognormal LMM..j )k. Vk (t)) dWkd (t). . provided that the spot LIBOR measure is chosen as reference for defining the basic volatility dynamics. but because the considered case deserves a thorough analysis due to its general features. Strikes and vols are expressed as differences from the respective ATM values. Wu and Zhang (2006) and Zhu (2007). and Z d = {Z1d . . correlated with Z . Rebonato (2007). Mercurio and Morini (2007) and Hagan and Lesniewski (2008). To this end. We then assume that the volatility process Vk evolves according to: dVk (t) = ak (t. FhD it + φk (t. . In this section. Dynamics (27) and (28) are general enough to include. we assume that the QTD -dynamics of each Lik (t) is given by dLik (t) = k X h=β(t) τhD dhLik . not only for the sake of details. the stochastic volatility Vk is an adapted process. Andersen and Andreasen (2002). i } compatible with We assume that we are given a set of times T = {0 < T0i . commonly referred to as SABR.

k)+1 We thus have the following.k) = (1{k>j} − 1{j>k} ) X h=min(j. FhD it dt 1 + τhD FhD (t) .1 Moving to a forward measure The standard change-of-numeraire technique. QDj ) T dhX. the drift of a given (continuous) process X changes according to Ti Drift(X. QTD ) h=β(t) In particular. x) = 0 bk (t.28 and Morini (2007) assume the following one-factor stochastic volatility dynamics dLik (t) = · · · dt + σk [Lik (t)]β V (t) dZkd (t) dV (t) = νV (t) dW d (t). QTD ) − j X h=β(t) = k X h=β(t) τhD dhLik . for the FRA rate Lik . are equivalent to those in the single-curve case. ln(BD /PD (·. when moving Ti from measure QTD to measure QDj . x) = σk xβ ψt (x) = x ak (t. we have Ti Drift(Lik . implies that. see also Section 6. FhD it = Drift(X. however. QDj ) = Drift(Lik . FhD it τhD dhLik . As we already explained in the lognormal case. FhD it dt 1 + τhD FhD (t) j X dhLik . FhD it τhD − dt dt 1 + τhD FhD (t) 1 + τhD FhD (t) h=β(t) max(j. (29) which amounts to choosing: φk (t. τhD dhLik . Tji ))it = − dt j X τhD dhX. and as such here omitted for brevity. β and ν are positive constants. QTD ) − dt 1 + τhD FhD (t) Drift(X. x) = νx where σk . the definition of a consistent LMM also requires the specification of the dynamics of the “discount” forward rates FkD and their related correlations. 7. Remarks on their possible specification will be provided at the end of the section. These dynamics.

. . ln τ it j=c+1 j PD (·. for each j. . Tiij ).d D ) c. FhD it + bk (t.d T dhX. the drift of process X changes according to Drift(X. .29 Ti Proposition 6 The dynamics of each Lik and Vk under the forward measure QDj are max(j..M .. τdS . TM Consider the annuity term c. ln(BD /CD )it = − dt  S Pd S PD (·.d T When moving from measure QD to QD . c. . there exists an index ij such that TjS = Tiij . . . FhD it + bk (t.Tj ) dhX. . with lognormal dynamics for the FRA rates. We can easily check that in the deterministic volatility case ak ≡ bk ≡ 0.k) dLik (t) X = (1{k>j} − 1{j>k} ) h=min(j. under QDk : dLik (t) = φk (t.j )k. 7. .d CD (t) = d X τjS PD (t. . ZM } is an M -dimensional QDj -Brownian motion with Ti j instantaneous correlation matrix (ρk. S with corresponding year fractions τc+1 .2 Moving to a swap measure S Let us denote by Tc+1 . Vk (t)) dt − k X h=β(t) τhD dhVk . QTD ) . ..T i ) β(t)−1 T = Drift(X. j=c+1 which is the numeraire associated to the swap measure Qc. Vk (t)) dt − j X h=β(t) τhD dhLik . WM } is also a QDj Brownian motion. . . . and W j = {W1j .j=1. for each j. {T0i . TjS ) j=c+1 = d X τjS PD (t. . Lik (t))ψt (Vk (t)) dZkk (t) dVk (t) = ak (t. (30) reduce to the lognormal LMM dynamics (20).k)+1 dVk (t) = ak (t. Qc. Vk (t)) dWkk (t) 1 + τhD FhD (t) (31) Ti j where. Vk (t)) dWkj (t) 1 + τhD FhD (t) (30) Ti In particular. QD ) + dt Drift(X. . the fixed-leg payment time of a given forward swap rate. ..d D . Lik (t))ψt (Vk (t)) dZkj (t) 1 + τhD FhD (t) τhD dhVk . Z j = {Z1j . . Then. . and assume that each TjS belongs to T = i }. . TdS . FhD it + φk (t.

Tβ(t)−1 PD (t. Tβ(t)−1 ) S S j=c+1 τj PD (t. Tiij ) i ) PD (t. Tj ) = 1. TjS ) Pd j=c+1 S S j=c+1 τj PD (t. TjS )  τhD i = − dhX. Tiij )  τhD dFhD (t) i PD (t. Tβ(t)−1 ) 1 + τhD FhD (t) τhD dF D (t) 1 + τhD FhD (t) h τhD dF D (t) 1 + τhD FhD (t) h Hence. Tj ) j=c+1 = · · · dt + Pd d X i ) PD (t. TjS )  S d ln τj i ) PD (t. TjS )  it i PD (·. Tβ(t)−1 j=c+1 = · · · dt + Pd d X i PD (t. TjS ) PD (·. Tβ(t)−1 1 + τ F (t) h h j=c+1 τj PD (t. Tj ) j=c+1 h=β(t) Lik . so that k X h=β(t) d k X X τjS PD (t. Tβ(t)−1 ij Y d h=β(t) τjS − h=β(t) X Pd S S j=c+1 τj PD (t. Tj ) j=c+1 h=β(t) Noticing that d X τjS PD (t. Pd k h t D D D D S S 1 + τh Fh (t) 1 + τh Fh (t) j=c+1 τj PD (t. F i − dhL dhLik . Tj ) j=c+1 = · · · dt + Pd d X i ) PD (t. TjS ) i ) PD (t.30 We need to calculate d  X PD (t. Qc. Tj ) j=c+1 h=β(t) . FhD it P t i D D d S S ) PD (·. ln d  X τjS j=c+1 = k X h=β(t) PD (·. Drift(Lik . FhD it . FhD it Pd k h t D D S S 1 + τhD FhD (t) 1 + τ F (t) h h j=c+1 τj PD (t. TjS ) τhD τhD i D . F i = dhLik . Tiij ) 1 τhD FhD (t) PD (t. Tj ) h=β(t) j=c+1 d X = · · · dt − ij X τjS PD (t. TjS ) τhD τhD i D dhL . Tβ(t)−1 S S j=c+1 τj PD (t. TjS ) Pd S S j=c+1 τj PD (t. ln d  X τjS j=c+1 which.d D ) dt leads to = Drift(Lik . QTD ) dt + dhX. the instantaneous covariation between X and the log of the numeraires ratio is given by dhX. Tβ(t)−1 S S j=c+1 τj PD (t. Tj ) j=c+1 d X = · · · dt − τjS d τjS τjS d PD (t. for X = ij d X X τjS PD (t. Tj ) j=c+1 = · · · dt + Pd d X i PD (t. Tβ(t)−1 ) S S j=c+1 τj PD (t. Tj ) h=β(t) j=c+1 1+ ij X  ij τjS PD (t. Tβ(t)−1 ) ij d X X τjS PD (t.

1 + τhD FhD (t) Ti where W c. Fh it + bk (t.j )k.d now denotes a one-dimensional QDj -Brownian motion. . TjS ) Pd ij X S S j=c+1 τj PD (t. FRA rate and volatility dynamics (30) and (32) differ from their single-curve homologues because of the different covariance terms in the drifts.d D -Brownian motion with c. Vk (t)) dt − j X h=β(t) τh dhFk . (32) c.d (t).d = {Z1c. 7.d .d . Tj ) h=β(t) j=c+1 τhD dhVk . when i ≡ D..d c.k) dFk (t) = (1{k>j} − 1{j>k} ) X h=min(j.31 we immediately have the following. for each j..d where. Lik (t))ψt (Vk (t)) dZkc. . ZM } is an M -dimensional Qc. .d D are dLik (t) = d X max(ij . .M . and τh is the associated year fraction. Vk (t)) dWkc. FhD it 1 + τhD FhD (t) + bk (t. Proposition 7 The dynamics of each Lik and Vk under the swap measure Qc. the dynamics under the Tji -forward measure are max(j.d (t) d X dVk (t) =ak (t. Fh it + φk (t. Vk (t)) dWkj (t) 1 + τh Fh (t) (33) where. .3 Specifying the covariations in the drifts As we already noticed in the lognormal LMM case. the swap-measure dynamics of the volatility process under the specification (29) is given by dV (t) = − d X τjS PD (t. Fh denotes the common value of FhD and Lih . FhD it + νV (t) dW c.j=1. WM } is also a Ti QDj -Brownian motion.d instantaneous correlation matrix (ρk. For instance. for each h.. . Vk (t)) dt − τjS PD (t. TjS ) Pd j=c+1 ij X S S j=c+1 τj PD (t. Z c. TjS ) Pd j=c+1 S S j=c+1 τj PD (t. Tj ) h=β(t) τhD dhV.k)+1 dVk (t) = ak (t.k)+1 τhD dhLik .d (t). and W = {W1c.k) τjS PD (t. FhD it D D 1 + τh Fh (t) + φk (t. Fk (t))ψt (Vk (t)) dZkj (t) 1 + τh Fh (t) τh dhVk . . For instance. Tj ) X (1{k>ij } − 1{ij >k} ) h=min(ij .. . .

. however. we are free to specify the former dynamics almost independently from the latter. WM } rate Lik but with possibly different parameters. . However.22 Another convenient choice would be to set to zero the instantaneous covariations in (30) and (32). in the dynamics of FkD and VkD . we assumed lognormal dynamics for the forward rates FhD . Vk (t)) dWk (t) D D D where φD k . the dynamics of Lik and Vk would depend on rates FhD only through their instantaneous correlations with them.D d.D . where νk ’s are deterministic functions of time. in fact.D D D D dhFkD . 22 If the derivative’s payoff depends on swap rates. ZM T are (highly) correlated M -dimensional QD -Brownian motions. drift corrections that are equivalent to those we obtain in the single-curve case. a convenient choice is to assume. we see from (30) that we must replace the forward rate Fk with the FRA rate Lik and the forward rates Fh (second argument in the covariations) with FhD (also when h = k). so that it may be more sensible to assume similar QTD -dynamics for rates Lik and FhD : dFkD (t) = k X h=β(t) dVkD (t) τhD d. ak and bk have the same form of the corresponding functions for the FRA d.D . these choices may not be so realistic. namely the forward rates FhD and their instantaneous covariations with them. too. . Moving to a forward or a swap measure will produce. FhD it + φD k (t. From a practical point of view. ZhD it + bk (t.D } and {W1d. . and {Z1d. Vk (t)) dWkj (t) h=β(t) In this case. yield curves i and D are likely to move in a similar (highly correlated) manner. . When dealing with two distinct curves. In principle. ZhD it + φk (t. In Section 6.k) dLik (t) = (1{k>j} − 1{j>k} ) X τhD νh (t) dhLik . since it implies that max(j. .D D D D = aD k (t. For example. . to mimic the evolution of the given FRA rates. so that no simulation of the dynamics of FhD would be needed in the Monte Carlo pricing of a LIBOR dependent derivative. See also Remark 2. which therefore need to be modeled. forward rates FhD need to be simulated anyway. for each k.k)+1 dVk (t) = ak (t. Vk (t)) dt + bk (t. the dynamics of Lik and Vk would remain the same irrespective of the chosen (forward or swap) measure. . dFkD (t) = · · · dt + νk (t)[1 + τkD FkD (t)]dZkD (t). Lik (t))ψt (Vk (t)) dZkj (t) h=min(j. ψt . .32 Dynamics (33) are fully specified by the instantaneous covariance structure of forward rates and their volatilities. independently of the dynamics of FhD . These dynamics of FRA rates and volatilities depend on extra quantities. In this case. Fk (t))ψt (Vk (t)) dZk (t) 1 + τhD FhD (t) d. Vk (t)) dt − j X τhD νh (t) dhVk .

For instance. so that we can write p dV (t) = [κθ − (κ + η)V (t))] dt +  V (t) dW k (t).D (t) D D D k X D dV (t) = κ (θ − V (t)) dt − h=β(t) p τhD D D D dhV . θD . . Zhk. . Zhk.4 Option pricing All the stochastic-volatility LMMs mentioned above lead to closed-form formulas for caps and swaptions. we will show how to price caps and swaptions in the Wu and Zhang (2006) model under constant coefficients. we may set:  p V (t)V D (t) k X h=β(t) τhD σkD FkD (t) dhW k . . F V D (t) dW k.33 7.D {Z1k. and {Z1k . F i +  t h 1 + τhD FhD (t) p dFkD (t) = σkD FkD (t) V D (t) dZkk. being already different in the basic single-curve case. . the volatility’s drift term produced by the measure change. whereas W k and W k. in different ways. Assuming single-factor stochastic-volatility dynamics of Heston (1993) type. . . the derivation procedures will be different. Ti Since the dynamics of V are (approximately) of square-root type also under QDk . κD . In particular. .D (t) i +  t h D D 1 + τh Fh (t) k where σk . the dynamics of V are more explicitly given by dV (t) = κ(θ − V (t)) dt −  p V (t)V D (t) k X h=β(t) p τhD σhD FhD (t) k. However. κ. Our double-curve setting allows for extensions of these formulas under each of these models. As an example. D are positive constants.D are Ti one-dimensional QDk -Brownian motions.D . see also Mercurio and Moreni (2006). ZM } and i T k. we can then price the caplet by means of Heston (1993) option formula. .D it D D 1 + τh Fh (t) p k V D (0) X τhD σhD FhD (0) ≈ V (t) p dhW k . we can approximate.D it =: ηV (t) dt D D V (0) h=1 1 + τh Fh (0) where the last equality defines the constant η parameter. the evoluTi tions under the forward measure QDk are: p dLik (t) = σk Lik (t) V (t) dZkk (t) dV (t) = κ(θ − V (t)) dt − k X h=β(t) p τhD D V (t) dW k (t) dhV. ZM } are M -dimensional QDk -Brownian motions.D k dhW .  and σkD . Z i +  V (t) dW k (t) t h 1 + τhD FhD (t) i To derive an analytical formula for the caplet paying [Lik (Tk−1 ) − K]+ at time Tki . θ. .

d D -dynamics of volatility V is more involved.b.d for some Qc. whereas the correlation between W k and Zhk.b. 8 Conclusions We have started by describing the change in value of the market interest rate quotes.b.d i = νSa.k ki . ν := t 2 [S (0)] a.d (t) V (t) dZa. and where v u b b uX X Li (0)Lih (0) ωk (0)ωh (0)σk σh ρh. meaning that no further approximation would be required. we can however freeze the weights ωk at their time-0 value. the dynamics of Sa.d k=a+1 p c.d (t) ≈ b X ωk (0)Lik (t). we noticed that once-compatible rates . This assumption is more innocuous than in the single-curve case. so as to be able to apply Heston’s (1993) option pricing formula also in the swaption case.d . a bit trickier.d (t) under its associated swap measure is (approximately) similar to that of each FRA rate under the corresponding forward measure.b.b.d (t) c.c. moreover. as usual. since the dynamics of V does not change as a consequence of the measure change. see equation (32).d (t) V (t) S (0) a.d k=a+1 h=a+1 i Therefore. The pricing of swaptions is.c. k=a+1 obtaining the following dynamics under the swap measure Qc.d (t) b X p ωk (0)σk Lik (t) dZkc.b.D .c. i Sa. As a major stylized fact.d (t) ≈ V (t) k=a+1 b X p Li (0) i ωk (0)σk i k dZkc. In fact. the correlation between Brownian motions W k and Zkk can be arbitrary and used for calibration of the market caplet skew.c. We notice once again that setting to zero the instantaneous correlation between V and the forward rates FhD would give a zero drift correction. As already done in the lognormal case.34 Remark 8 We notice that an exact formula can be obtained by simply setting to zero the correlation between the Brownian motions W k and Zhk. we can again resort to freezing techniques and derive approximated dynamics of square-root type.c.b.c.c.c. In the double-curve case.D can be conveniently assumed to be zero without creating any contradiction (unless the whole correlation matrix is not positive definite). which occurred since August 2007. The Qc.b. we must face the further complication that swap rates depend also on forward rates FkD and not only on FRA rates Lik .c.b.d (t) ≈ Sa.d D -Brownian motion Za. However.d D : i dSa.

(2008) Two Curves. a similar assumption has to be made on the times defining the fixed and floating legs of the forward swaps in question. Scandolo (2007) Liquidity Risk Theory and Coherent Measures of Risk. Working Paper. Mercurio. In this article. Interest-Rate Models: Theory and Practice. . . considering a general dynamics that contains as a particular case all stochastic volatility LMMs known in the financial literature. References [1] Gatarek. The market model of interest rate dynamics. D. C. Available online at: http://papers. With Smile. and W. D. Working Paper. Available online at: www. [3] Andersen. Springer Finance. 127–154.. producing a clear segmentation of market rates. one for each tenor considered. [7] Brigo. M. TM } is included in the other. edited by F. When pricing payoffs depending on swap rates.frankfurt-school. W. and M. 6 and 12 months. This is usually the time structure {T0i . Mathematical Finance. . L. .cfm?abstract_id=1048322 [2] Ametrano. 3. Mercurio (2006). F. . [5] Boenkost. We have then extended the basic lognormal LMM and derived its dynamics under the relevant measure changes. Musiela (1997). and J. HfBBusiness School of Finance & Management. Volatile volatilities. December.. . Andreasen (2002). The pricing formulas for caps and swaptions result in a simple modification of the corresponding Black formulas used by the market in the single-curve setting. If two LIBOR curves i and j enters a given payoff at the same time. since the tenors that are typically considered in the market are 1. and G. Practitioners tackled the issue by building different yield curves for different rate tenors. . The analysis with two distinct yield curves can be extended to allow for the simultaneous presence of more “growth” curves. 163-168. Inflation and Credit. Risk. This is fundamental if we have to price a contract that depends on different LIBOR or swap rates. Cross currency swap valuation. we just have to assume that either j i } or {T0j . and M. A.35 began to diverge [6] Brace. [4] Bianchetti. We have concluded by introducing stochastic volatility. Bianchetti (2008) in Modelling Interest Rates: Latest Advances for Derivatives Pricing.ssrn. Risk Books. 7. and F. . . One Price: Pricing & Hedging Interest Rate Derivatives Using Different Yield Curves for Discounting and Forwarding. Schmidt (2005). TM case in practice. we have shown how to price the main (linear and plain vanilla) interest rate derivatives under the assumption of two distinct curves for generating future LIBOR rates and for discounting.

. A Multi-Quality Model of Interest Rates. and Other Derivatives (5th Edition). [9] Fruchard. Review of Financial Studies 19.. [16] Jamshidian. November. 92-97.. F. Tanaka and T. (2005) Stochastic volatility model with time-dependent skew. (1996) Sorting out Swaptions.. (2007) A time Homogeneous. [23] Rebonato. SABR-Consistent Extension of the LMM. Prentice Hall. 12(2). Jarrow. Futures. 409-430. Risk 20. The puzzle in the interest rate curve: counterparty risk? Preprint. (1997) LIBOR and Swap Market Models and Measures.. Finance and Stochastics 1. Sandmann and D. F. and A. P. P.cfm?abstract_id=1018026 [20] Miltersen. Moreni (2006).lesniewski. Risk. [11] Hagan. Sondermann (1997). The Review of Financial Studies 6. [13] Heston. September. Kumar. D. Warachka (2006). [22] Piterbarg. The Journal of Finance 52. Protter and M. 84-108. Lesniewski (2008). R. Willems (1995). F. K.R.pdf [12] Henry-Labord`ere. 493-529.. Risk 19(3). V. (2007) Combining the SABR and LMM Models. Basis for change. U. Woodward. F. P. Quantitative Finance. Risk. LIBOR market model with SABR style stochastic volatility. Milan. 7075. [15] Jamshidian. P. Applied Mathematical Finance.. and N.ssrn. 293-330.S. Wong (2008). 59-60.36 [8] Cetin. [21] Morini. M. 327-343. J. K. K. (1993) A Closed-Form Solution for Options with Stochastic Volatility with Applications to Bond and Currency Options. S. Wilmott magazine. [18] Mercurio. [14] Hull. Available online at: http://papers. Closed Form Solutions for Term Structure Derivatives with Log-Normal Interest Rates. C. D. and M. (2002) Options.L. . [19] Mercurio.. No-Arbitrage dynamics for a tractable SABR term structure Libor Model. 147-185. Pricing Options in an Extended Black Scholes Economy with Illiquidity: Theory and Empirical Evidence. Risk October.E. Zammouri and E. (2002) Managing Smile Risk. Lesniewski. 102-107. March. Banca IMI. [10] Hagan.. Inflation with a smile. R. Available online at: http://www. M (2008). A. [17] Kijima. Morini (2007).

Zhang (2006). LIBOR market model with stochastic volatility. L.cfm?abstract_id=955352 . An Extended Libor Market Model With Nested Stochastic Volatility Dynamics. 2.37 [24] Sch¨onbucher.cfm?abstract_id=261051 [25] Wu. 2. (2000) A Libor Market Model with Default Risk. Available online at: http://papers. 199227. (2007). [26] Zhu.ssrn. P.ssrn. Journal of Industrial and Management Optimization. Available online at: http://papers. and F. J.