Two Curves, One Price: Pricing & Hedging Interest Rate Derivatives Using Dierent Yield Curves for

Discounting and Forwarding

Marco Bianchetti Risk Management, Market Risk, Pricing and Financial Modeling Banca Intesa Sanpaolo, piazza P. Ferrari 10, 20121 Milan, Italy Version 1.4, Jan. 29th, 2009

Abstract In this paper we revisit the problem of pricing and hedging plain vanilla singlecurrency interest rate derivatives using dierent yield curves for market coherent estimation of discount factors and forward rates with dierent underlying rate tenors (e.g. Euribor 3 months, 6 months,.etc.). Within such double-curve-single-currency framework, adopted by the market after the liquidity crisis started in summer 2007, standard single-curve no arbitrage relations are no longer valid and can be formally recovered through the introduction of a basis adjustment. Numerical results show that the basis adjustment curves may display, in trouble market times, an oscillating micro-term structure, strongly dependent on the quality of the bootstrapping. Such shapes may induce appreciable eects on the price of interest rate instruments, in particular when people switches from the single-curve towards the double-curve framework. Recurring to the foreign-currency analogy we also derive the no arbitrage doublecurve market-like formulas for basic plain vanilla interest rate derivatives, FRA, swaps, cap/oors and swaptions in particular. These expressions include an extra quanto adjustment term typical of cross-currency derivatives, naturally originated by the change between the numeraires associated to the two yield curves, that carries on a volatility and correlation dependence. Numerical scenarios conrm that such correction can be non-negligible, thus making the market prices, in principle,
The author acknowledges fruitful discussions with M. De Prato, C. Ma, F. Mercurio, N. Moreni, M. Morini, M. Pucci and with many colleagues in the Risk Management. A particular mention is due to F. Ametrano and to the QuantLib community for the open-source developements used to compute the numerical results reported in the paper. The opinions expressed here are solely of the author and do not represent in any way those of his employer.

not arbitrage free. In practice arbitrage opportunities are hidden by the market incompleteness. JEL Classications: E45, G13. Keywords: liquidity, crisis, yield curve, forward curve, discount curve, pricing, hedging, interest rate derivatives, FRAs, swaps, basis swaps, caps, oors, swaptions, basis adjustment, quanto adjustment, measure changes, no arbitrage.

Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 Single-Curve Pricing & Hedging Interest-Rate Derivatives 1.2 From Single to Double-Curve Paradigm . . . . . . . . . . 2 Double-Curve Framework, No Arbitrage and Basis Adjustment 2.1 General Assumptions . . . . . . . . . . . . . . . . . . . . . 2.2 Pricing Procedure . . . . . . . . . . . . . . . . . . . . . . 2.3 No Arbitrage Revisited and Basis Adjustment . . . . . . . 3 Foreign-Currency Analogy and Quanto Adjustment . . . . . . . 3.1 Forward Rates . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Swap Rates . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Double-Curve Pricing & Hedging Interest Rate Derivatives . . . 4.1 Pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Hedging . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 3 4 7 7 9 10 12 13 16 19 19 21 22

1. Introduction
One of the many consequences of the liquidity crisis started in the second half of 2007 has been a strong increase of the basis spreads quoted on the market between single-currency interest rate instruments, swaps in particular, characterized by dierent underlying rate tenors (e.g. Euribor3M1 , Euribor6M, etc.), reecting the increased liquidity risk and the corresponding preference of nancial institutions for receiving payments with higher frequency (quarterly instead of semi-annualy, for instance). Such asymmetry has induced a sort of segmentation of the interest rate market into sub-areas, mainly corresponding to instruments with 1M, 3M, 6M, 12M underlying rate tenors. Each area is characterized, in principle, by its own internal dynamic, reecting the dierent views and interests of the market players. In gs. 1.1, 1.2 we show a snapshot of the market quotations as of 30 Sep. 2008 (a sensitive date for quarterly balance sheet results of nancial institutions and industries)
1 Euro Interbank Oered Rate, the rate at which euro interbank term deposits within the euro zone are oered by one prime bank to another prime bank (see e.g. www.euribor.org).

for six basis swap curves corresponding to the four Euribor tenors 1M, 3M, 6M, 12M. As one can see, the basis spreads are monotonically decreasing from over 100 to around 4 basis points. There is neither way nor any good reason to ignore such quotations in a market-coherent pricing framework of interest rate derivatives. We stress that the present market situation described above is nothing else that a new equilibrium conguration determined by the pressure of the increased illiquidity force, that enlarges well known eects hystorically very small and traditionally neglected before the crisis. 1.1. Single-Curve Pricing & Hedging Interest-Rate Derivatives Such evolution of the nancial markets has triggered a general reection about the methodology used to price and hedge interest rate derivatives, namely those nancial instruments whose price depends on the present value of future interest rate-linked cashows. The pre-crisis standard market practice (which does not automatically mean good practice) can be summarized in the following procedure (see e.g. refs. [1]-[4]): 1. select one nite set of the most convenient (e.g. liquid) vanilla interest rate instruments traded in real time on the market with increasing maturities; for instance, a very common choice in the EUR market is a combination of short-term EUR deposit, medium-term Futures on Euribor3M and medium-long-term swaps on Euribor6M; 2. build one yield curve using the selected instruments plus a set of bootstrapping rules (e.g. pillars, priorities, interpolation, etc.); 3. compute on the same curve forward rates, cashows2 , discount factors and work out the prices by summing up the discounted cashows; 4. compute the delta sensitivity and hedge the resulting delta risk using the suggested amounts (hedge ratios) of the same set of vanillas. For instance, a 5.5Y maturity EUR oating swap leg on Euribor1M (not directly quoted on the market) is commonly priced using discount factors and forward rates calculated on the same depo-Futures-swap curve cited above. The corresponding delta sensitivity is calculated by shoking one by one the curve pillars and the resulting delta risk is hedged using the suggested amounts (hedge ratios) of 5Y and 6Y Euribor6M swaps3 .
within the present context of interest rate derivatives we focus in particular on forward rate dependent cashows. See also eq. 2.11. 3 we refer here to the case of local yield curve bootstrapping methods, for which there are no sensitivity delocalization eect (see refs. [1], [2]).
2

We stress that this is a single-currency-single-curve approach, in that a unique curve is built and used to price and hedge any interest rate derivative on a given currency. Thinking in terms of more fundamental variables, e.g. the short rate, this is equivalent to assume that there exist a unique fundamental underlying short rate process able to model and explain the whole term structure of interest rates of any tenor. It is also a relative pricing approach, because both the price and the hedge of a derivative are calculated relatively to a set of vanillas quoted on the market. We notice also that the procedure is not strictly guaranteed to be arbitrage-free, because discount factors and forward rates obtained through interpolation are, in general, not necessarily consistent with the no arbitrage condition; in practice bid-ask spreads and transaction costs virtually hide any arbitrage possibility. Finally, we stress that the rst key point in the procedure above is much more a matter of art than of science, because there is not an unique nancially sound choice of bootstrapping instruments and, in principle, none is better than the others. The methodology described above can be extended, in principle, to more complicated cases, in particular when a model of the underlying interest rate evolution is used to calculate the future dynamic of the yield curve and the expected cashows. The volatility and (eventually) correlation dependence carried by the model implies, in principle, the bootstrapping of a variance/covariance matrix (two or even three dimensional) and hedging the corresponding sensitivities (vega and rho) using volatility and correlation dependent vanilla market instruments. In practice just a small subset of such quotations is available on the market, and thus only some portions of the variance/covariance matrix can be extracted from the market. In this note we will focus only on the basic matter of yield curves and forget the volatility/correlation dimensions. 1.2. From Single to Double-Curve Paradigm Unfortunately, the pre-crisis approach outlined above is no longer consistent, at least in this simple formulation, with the present market conguration. First, it does not take into account the market information carried by the basis swap spreads, now much larger than in the past and no longer negligible. Second, it does not take into account that the interest rate market is segmentated into sub-areas corresponding to instruments with dierent underlying rate tenors, characterized, in principle, by dierent dynamics (e.g. short rate processes). Thus, pricing and hedging an interest rate derivative on a single yield curve mixing dierent underlying rate tenors can lead to dirty results, incorporating the dierent dynamics, and eventually the inconsistencies, of dierent market areas, making prices and hedge ratios less stable and more dicult to interpret. On the other side, the more the vanillas and the derivative share the same homogeneous underlying rate, the better should be the relative pricing and the hedging. 4

Third, by no arbitrage, discounting must be univocal: two identical future cashows of whatever origin must display the same present value; hence we need an unique discounting curve. The market practice has thus evolved to take into account the new market informations cited above, that translate into the additional requirement of homogeneity: as far as possible, interest rate derivatives with a given underlying rate tenor should be priced and hedged using vanilla interest rate market instruments with the same underlying. The corresponding pricing procedure will be formalized and justied in sec. 2.2; we summarize here the following modied working procedure: 1. build one discounting curve using the preferred procedure; 2. select multiple separated sets of vanilla interest rate instruments traded in real time on the market with increasing maturities, each set homogeneous in the underlying rate (typically with 1M, 3M, 6M, 12M tenors); 3. build multiple separated forwarding curves using the selected instruments plus their bootstrapping rules; 4. compute on each forwarding curve the forward rates and the corresponding cashows relevant for pricing derivatives on the same underlying; 5. compute the corresponding discount factors using the discounting curve and work out prices by summing up the discounted cashows; 6. compute the delta sensitivity and hedge the resulting delta risk using the suggested amounts (hedge ratios) of the corresponding set of vanillas. For instance, the 5.5Y oating swap leg cited in the previous section should be priced using Euribor1M forward rates calculated on an pure 1M forwarding curve, bootstrapped only on Euribor1M vanillas, plus discount factors calculated on the discounting curve. The corresponding delta sensitivity should be calculated by shoking one by one the pillars of both yield curves, and the resulting delta risk hedged using the suggested amounts (hedge ratios) of 5Y and 6Y Euribor1M swaps plus the suggested amounts of 5Y and 6Y instruments from the discounting curve (see sec. 4.2 for more details about the hedging procedure). The improved approach described above is more consistent with the present market situation, but - there is no free lunch - it does demand much more additional eorts. First, the discounting curve clearly plays a special and fundamental role, and must be built with particular care. This pre-crisis obvious step has become, in the present market situation, a very subtle and controversial point, that would require a whole paper in itself. In fact, while the forwarding curves construction is driven by the underlying 5

rate tenor homogeneity principle, for which there is (now) a general market consensus, there is no longer general consensus for the discounting curve construction. At least two dierent practices can be encountered on the market: a) the old pre-crisis approach (e.g. the depo, Futures and swap curve cited before), that can be justied with the principle of maximum liquidity (plus a little of inertia), and b) the Eonia4 curve, justied with no risky or collateralized counterparties, and by increasing liquidity (see e.g. the discussion in ref. [5]). Second, building multiple curves requires multiple quotations: much more interest rate bostrapping instruments must be considered (deposits, Futures, swaps, basis swaps, FRAs, etc.), which are available on the market with dierent degrees of liquidity and can display transitory inconsistencies. Third, non trivial interpolation algorithms are crucial to produce smooth forward curves (see e.g. refs. [1]-[3]). Fourth, multiple bootstrapping instruments implies multiple sensitivities, so hedging becomes more complicated. Last but not least, pricing libraries, platforms, reports, etc. must be extendend, congured, tested and released to manage multiple and separated yield curves for forwarding and discounting, not a trivial task for quants, developers and IT people. In this paper we assume the existence of a convenient methodology (described in detail in ref. [6]) for bootstrapping a discounting curve plus multiple forwarding curves characterized by dierent underlying rate tenors (Euribor1M, 3M, 6M and 12M), and we focus on the consequences for pricing and hedging interest rate derivatives. In particular in sec. 2 we x the notation, we revisit some general concept of standard, no arbitrage single-curve pricing and we formulate the principles for double-curve pricing, showing how no arbitrage can be formally recovered with the introduction of a basis adjustment. In sec. 3 we use the foreign-currency analogy to derive a single-currency version of the quanto adjustment (typical of cross-currency derivatives) to be applied to forward rates. In sec. 4 we derive the no arbitrage double-curve pricing expression for basic single-currency interest rate derivatives, zero coupon bonds, FRA, swaps and cap/oor options in particular. Finally, in sec. 5 we summarize the conclusions. Some of these topics have been approached also in other papers (see e.g. refs. [7][12]). In particular W. Boenkost and W. Schmidt [9] discuss two methodologies for pricing cross-currency basis swaps, the rst of which (the actual pre-crisis common market practice), does coincide, once reduced to the single-currency case, with the double-curve pricing procedure described here. The concerns expressed by these authors, that such approach was not arbitrage free and not consistent with the pre-crisis singlecurve market practice for pricing single-currency swaps, have now been overcome by the market evolution towards a generalized double-curve pricing approach. Recently M. Kijima et al. [10] have extended the approach of ref. [9] to the case of three curves for
4 Euro OverNight Index Average, the rate computed as a weighted average of all overnight rates corresponding to unsecured lending transactions in the euro-zone interbank market (see e.g. www.euribor.org).

discount factors, swap forward rates and bond forward rates, bootstrapped using swaps, cross-currency swaps and governative bonds, respectively. Finally, simultaneously to the drafting of the present paper, M. Morini [11] is approaching the same problem in terms of counterparty risk, while F. Mercurio [12] is rebuilding the whole theory from scratch in terms of modied Libor Market Model. The present work is complementary to those cited before in the sense that: a) we adopt a practitioners perspective, discussing in detail the current market practices and keeping things as simple as possible; b) we explain why the market has evolved from a single-curve towards a double-curve pricing framework in the case of singlecurrency interest rate instruments (to our knowledge, this is discussed in detail only in ref. [12]); c) we proof how non-arbitrage is broken-up and can be recovered in terms of basis adjustment; d) we use the foreign-currency analogy to derive pricing formulas for basic interest rate derivatives including the quanto adjustment arising from the change between the numeraires (or probability measures) naturally associated to forwarding and discounting curves; e) last but not least, we consistently keep together both pricing and hedging issues, whose intimate connection is at the heart of quantitative nance, trading and risk management.

2. Double-Curve Framework, No Arbitrage and Basis Adjustment

2.1. General Assumptions We start by postulating the existence of two dierent interest rate markets, denoted by Mx , x = {d, f }, characterized by the same currency and by two distinct bank accounts Bx and yield curves {x in the form of a continuous term structure of discount factors, {x = {T Px (t0 , T ) , T t0 } , (2.1)

where t0 is the (common) reference date of the curves (e.g. settlement date, or today) and Px (t, T ) denotes the price at time t t0 of the x -zero coupon bond for maturity T , such that Px (T, T ) = 1. Next we postulate the usual no arbitrage relation in each market Mx , Px (t, T2 ) = Px (t, T1 ) Px (t, T1 , T2 ) , (2.2)

where t T1 < T2 and Px (t, T1 , T2 ) denotes the forward discount factor from time T2 to time T1 , prevailing at time t. The nancial meaning of expression 2.2 is that, in each market Mx , given a cashow of one unit of currency at time T2 , its corresponding value at time t < T2 must be the same both if we discount in one single step from T2 to t, using the discount factor Px (t, T2 ), and if we discount in two steps, rst from T2 to T1 , using the forward discount Px (t, T1 , T2 ) and then from T1 to t, using Px (t, T1 ). Denoting with 7

Fx (t; T1 , T2 ) the simple compounded forward rate associated to Px (t, T1 , T2 ), resetting at time T1 and covering the time interval [T1 ; T2 ], we have Px (t, T1 , T2 ) = 1 Px (t, T2 ) = , Px (t, T1 ) 1 + Fx (t; T1 , T2 ) x (T1 , T2 ) (2.3)

where x (T1 , T2 ) is the year fraction between times T1 and T2 with daycount dcx , from eq. 2.2 we obtain the familiar no arbitrage expression Fx (t; T1 , T2 ) = = 1 1 1 x (T1 , T2 ) Px (t, T1 , T2 ) Px (t, T1 ) Px (t, T2 ) . x (T1 , T2 ) Px (t, T2 )

(2.4)

Eq. 2.4 can be also derived (see e.g. ref. [13], par. 1.4) as the fair value condition at time t of the Forward Rate Agreement (FRA) contract with payo at maturity T2 given by FRA (T2 ; T1 , T2 , K, N ) = N x (T1 , T2 ) [Lx (T1 , T2 ) K] , 1 Px (T1 , T2 ) Lx (T1 , T2 ) = x (T1 , T2 ) Px (T1 , T2 ) (2.5) (2.6)

where N is the nominal amount, Lx (T1 , T2 , dcx ) is the T1 -spot Euribor rate and K the (simply compounded) strike rate (sharing the same daycount convention for simplicity). Introducing expectations we have, t T1 < T2 , FRA (t; T1 , T2 , K, N ) = Px (t, T2 ) EQx [FRA (T2 ; T1 , T2 , K, N )] t T Qx2 = N Px (t, T2 ) x (T1 , T2 ) Et [Lx (T1 , T2 )] K = N Px (t, T2 ) x (T1 , T2 ) [Fx (t; T1 , T2 ) K] ,
T2

(2.7)

where QT2 denotes the Mx -T2 -forward measure, EQ [.] denotes the expectation at time x t t w.r.t. measure Q and ltration Ft (encoding the market information available up to time t), and we have used the standard martingale property of forward rates Fx (t; T1 , T2 ) = EQx [Fx (T1 ; T1 , T2 )] = EQx [Lx (T1 , T2 )] t t holding in each interest rate market Mx . 8
T2 T2

(2.8)

2.2. Pricing Procedure Now we consider any derivative written on a single underlying interest rate with n future coupons with payos = { 1 , ..., n }, generating n cashows c = {c1 , ..., cn } at future dates T = {T1 , ..., Tn }, t < T1 < ... < Tn . Following the discussion in sec. 1.2, we postulate the following generalized, double-curve-single-currency pricing procedure: 1. assume {d as the discounting curve and {f as the forwarding curve; 2. calculate any relevant spot/forward rate on the forwarding curve {f as Lf (t, Ti ) = Ff (t; Ti1 , Ti ) = 1 Pf (t, Ti ) , t < Ti , f (t, Ti ) Pf (t, Ti ) Pf (t, Ti1 ) Pf (t, Ti ) , t Ti1 < Ti , f (Ti1 , Ti ) Pf (t, Ti ) (2.9) (2.10)

3. calculate cashows ci , i = 1, ..., n, as expectations of the i-th coupon payo i with respect to the discounting Ti -forward measure QTi , d ci = c (t, Ti , i ) = Et
Qd i
T

[ i ] ;

(2.11)

4. calculate the price at time t by discounting each cashow c (t, Ti , i ) using the corresponding discount factor Pd (t, Ti ) obtained from the discounting curve {d and summing up, (t; T) = =
n X i=1 n X i=1

Pd (t, Ti ) c (t, Ti , i ) Pd (t, Ti ) Et

Qd i
T

[ i ] .

(2.12)

Notice that steps 3 and 4 above have been formulated in terms of the particular pricing measure QTi associated to the numeraire Pd (t, Ti ). This is convenient in our d context because it emphasizes that discounting must be associated to the discounting curve. Obviously they can be reformulated in terms of any other equivalent measure associated to dierent numeraires. If we apply the paradigm above to the FRA case, we obtain that the single-curve FRA price in eq. 2.7 traslates into the following generalized, double-curve expression T Q 2 FRA (t; T1 , T2 , K, N ) = N Pd (t, T2 ) f (T1 , T2 ) Et d [Ff (T1 ; T1 , T2 )] K . (2.13) We will compute such expectation in par. 3. 9

2.3. No Arbitrage Revisited and Basis Adjustment The rst consequence of the assumptions above is that, clearly, standard single-curve no arbitrage relations such as eq. 2.4 are broken up, being Pd (t, T1 , T2 ) = 1 Pd (t, T2 ) = Pd (t, T1 ) 1 + Fd (t; T1 , T2 ) d (T1 , T2 ) Pf (t, T2 ) 1 = = Pf (t, T1 , T2 ) . (2.14) 6= 1 + Ff (t; T1 , T2 ) f (T1 , T2 ) Pf (t, T1 )

Clearly, no arbitrage is immediately recovered by postulating the following generalized double-curve no arbitrage relation Pd (t, T1 , T2 ) = = Pd (t, T2 ) Pd (t, T1 ) 1 1 + Ff (t; T1 , T2 ) BAf d (t, T1 , T2 ) f (T1 , T2 ) , (2.15)

or the equivalent simple transformation rule for forward rates Fd (t; T1 , T2 ) = Ff (t; T1 , T2 ) BAf d (t, T1 , T2 ) . (2.16)

We call the conversion factor BAf d (t, T1 , T2 ) in eqs. 2.15-2.16 (forward) basis adjustment. From eq. 2.16 we can express it as a ratio between forward rates or, equivalently, in terms of discount factors from {d and {f curves as BAf d (t, T1 , T2 ) = = Fd (t; T1 , T2 ) Ff (t; T1 , T2 ) f (T1 , T2 ) Pf (t, T2 ) Pd (t, T1 ) Pd (t, T2 ) . d (T1 , T2 ) Pd (t, T2 ) Pf (t, T1 ) Pf (t, T2 )

(2.17)

Notice that if {d = {f we recover the single-curve case BAf d (t, T1 , T2 ) = 1. In eq. 2.16 we have chosen a multiplicative denition of the basis adjustment. Obviously the alternative additive denition is completely equivalent BA0 d (t, T1 , T2 ) = Fd (t; T1 , T2 ) Ff (t; T1 , T2 ) f =

Pd (t, T1 ) Pd (t, T2 ) Pf (t, T1 ) Pf (t, T2 ) d (T1 , T2 ) Pd (t, T2 ) f (T1 , T2 ) Pf (t, T2 ) = Ff (t; T1 , T2 ) [BAf d (t, T1 , T2 ) 1] .

(2.18)

The latter is more useful for comparisons with the market basis spreads of gs. 1.1-1.2. 10

We stress that the basis adjustment in eqs. 2.17-2.18 is a straightforward consequence of the assumptions above, essentially the existence of two curves and no arbitrage. In practice its value depends on the basis spread between the quotations of the two sets of vanilla instruments used in the bootstrapping of the two curves {d and {f . The advantage of expressions 2.17, 2.18 is that they allows for a direct computation of the basis adjustment between forward rates for any time interval [T1 , T2 ], which is the relevant quantity for pricing and hedging interest rate derivatives. On the other side, the limit of expression 2.17-2.18 is that they reects the statical 5 dierences between the two interest rate markets Md , Mf carried by the two curves {d , {f , but they are completely independent of the interest rate dynamics in Md and Mf . Notice also that, in principle, we can also use our approach to bootstrap a new yield curve from a given yield curve plus a given basis adjustment. Inverting eq. 2.17 we obtain the following recursive relations Pd,i = = Pf,i = = f,i Pf,i Pd,i1 d,i [Pf,i1 Pf,i ] BAf d,i + f,i Pf,i1 f,i Pf,i Pd,i1 , d,i [Pf,i1 Pf,i ] + d,i f,i Pf,i BA0 d,i + f,i Pf,i1 f d,i Pd,i BAf d,i Pf,i1 f,i [Pd,i1 Pd,i ] + d,i Pd,i1 BAf d,i d,i Pd,i Pf,i1 , d,i Pd,i + f,i [Pd,i1 Pd,i ] d,i f,i Pd,i BA0 d,i f

(2.19)

(2.20)

where we have shortened the notation by putting x (Ti1 , Ti ) = x,i , Px (t, Ti ) = Px,i , BAf d (t, Ti1 , Ti ) = BAf d,i . Given the yield curve up to step Px,i1 plus the basis adjustment for the step i 1 i, the equations above can be used to obtain the next step Px,i . We now discuss a numerical example of the basis adjustment in a realistic market situation. We consider the four interest rate underlyings I = {I1M , I3M , I6M , I12M }, where I = Euribor index, and we bootstrap from market data ve separated yield curves { = {{d , {I1M , {I2M , {I6M , {I12M }, using the rst one for discounting and the f f f f others for forwarding. We follow the methodology described in ref. [6] and we use the corresponding open-source development in the QuantLib framework [14]. The discounting curve {d is built following a typical pre-crisis standard recipe from the most liquid deposit, 3M Futures and 6M swap contracts; the other curves are built from mixings of depos, FRAs, Futures, swaps and basis swaps with homogeneous underlying rate tenors; a smooth algorithm (monotonic cubic spline on log discounts) is used for interpolations6 .
5 we remind that the discount factors in eqs. 2.17-2.17 are calculated on the curves {d , {f following the recipe described in sec. 2.2, not using any dynamical model for the evolution of the rates. 6 technicalities of the curves construction are not crucial in the present context.

11

In g. 2.1 we plot the 6M-tenor forward rates calculated on {d and {I6M as of 15 f Sept. 2008, end of day. This is an highly stressed market period, just at the Lehman default and after the Fannie Mae and Freddie Mac federal takeover (Sep. 8, 2008). The eects of such main market events are clearly visible in the crazy roller-coaster look-up of the curves. The small dierences in the two short-term structures derive from the use of dierent market instruments in the two bootstrappings, while the medium and long-term similarity is due to the common use of 6M swap quotes. Similar patterns are observed also in the other 1M, 3M, 12M curves (not reported here). In g. 2.2 (upper panels) we plot the term structure of the four corresponding multiplicative basis adjustment curves calculated through eq. 2.17. In the lower panels we also plot the additive basis ajustment given by eq. 2.18. The higher short-term basis adjustments (left panels) are due to the higher short-term market basis spreads (see Figs. 1.1-1.2). We observe in particular that the medium-long-term {I6M {d f basis (dash-dotted green lines in the right panels) are close to 1 and 0, respectively, as expected from the common use of 6M swaps. A similar, but less evident, behaviour is found in the short-term {I3M {d basis (continuous blue line in the left panels), as f expected from the common 3M Futures and the uncommon deposits. The two remaining basis curves {I1M {d and {I12M {d are generally far from 1 or 0 because of dierent f f bootstrapping instruments. Obviously such details depend on our arbitrary choice of the discounting curve. Overall, we notice that all the basis curves {x {d reveal a complex micro-term f structure, not present in the smooth and monotonic basis swaps market quotes of gs. 1.1-1.2, essentially due to an amplication eect of small local dierences between the {d and {x yield curves. In general, such richer term structure is a very sensitive test f of the quality of the bootstrapping procedure (interpolation in particular), and also an indicator of the tiny, but observable, dierences between dierent interest rate market areas. Obviously these causes may have appreciable eects on the price of similar interest rate instruments. We can appreciate in g. 2.3 the perverse eects of a non smooth bootstrapping (linear interpolation on zero rates, a common market practice). The angular points in the lower panel clearly show the inadequacy of the bootstrap, but the very strong oscillations in the (additive) basis adjustment in the upper panel (notice the dierent scales w.r.t. g. 2.2, lower panels) allows to further appreciate the unnatural dierences induced in similar forward instruments priced on the two curves.

3. Foreign-Currency Analogy and Quanto Adjustment

We come now to the problem of calculating expectations as in eq. 2.12 in general and 2.13 in particular. This will involve the dynamical properties of the two interest rate markets Md and Mf , or, in other words, will require to model the dynamics for the interest rates in Md and Mf . This task is easily accomplished by recurring to the 12

natural analogy with cross-currency derivatives. Going back to the beginning of sec. 2, we can identify Md and Mf with the domestic and foreign markets, {d and {f with the corresponding curves, and the bank accounts Bd (t), Bf (t) with the corresponding currencies, respectively7 . Within this framework, we can recognize on the r.h.s of eq. 2.15 the forward discount factor from time T2 to time T1 expressed in domestic currency, and on the r.h.s. of eq. 2.13 the expectation of the foreign forward rate w.r.t the domestic forward measure. Hence, the computation of such expectation must involve the quanto adjustment commonly encountered in the pricing of cross-currency derivatives. The derivation of such adjustment can be found in standard textbooks. Anyway, in order to fully appreciate the parallel with the present double-curve-single-currency case, it is useful to run through it once again. In particular, we will adapt to the present context the discussion found in ref. [13], chs. 2.9 and 14.4. 3.1. Forward Rates In the doublecurve-double-currency case, no arbitrage requires the existence at any time t0 t T of a spot and a forward exchange rate between equivalent amounts of money in the two currencies such that cd (t) = xf d (t) cf (t) , Xf d (t, T ) Pd (t, T ) = xf d (t) Pf (t, T ) , (3.1) (3.2)

where the subscripts f and d stand for foreign and domestic, cd (t) is any cashow (amount of money) at time t in units of domestic-currency and cf (t) is the corresponding cashow at time t (the corresponding amount of money) in units of foreign currency. Obviously Xf d (t, T ) xf d (t) for t T . Expression 3.2 is a direct consequence of no arbitrage. This can be understood with the aid of g. 3.1: starting from top right corner in the time vs currency/yield curve plane with an unitary cashow at time T > t in foreign currency, we can either move along path A by discounting at time t on curve {f using Pf (t, T ) and then by changing into domestic currency units using the spot exchange rate xf d (t), ending up with xf d (t) Pf (t, T ) units of domestic currency; or, alternatively, we can follow path B by changing at time T into domestic currency units using the forward exchange rate Xf d (t, T ) and then by discounting on {d using Pd (t, T ), ending up with Xf d (t, T ) Pd (t, T ) units of domestic currency. Both paths stop at bottom left corner, hence eq. 3.2 must hold by no arbitrage. Our double-curve-single-currency case is immediately obtained from the discussion above by thinking to the subscripts f and d as shorthands for forwarding and discounting and by recognizing that, having a single currency, the spot exchange rate must collapse
7 notice the lucky notation: d stands either for discounting ordomestic and f for forwarding or foreign, respectively.

13

to 1. We thus have xf d (t) = 1, Pf (t, T ) . Xf d (t, T ) = Pd (t, T ) (3.3) (3.4)

Obviously for {d = {f we recover the single-currency, single-curve case Xf d (t, T ) = 1 T . The interpretation of the forward exchange rate in eq. 3.4 within this framework is straightforward: it is nothing else that the counterparty of the (forward) basis adjustment in eq. 2.16 for discount factors on the two yield curves {d and {f . They satisfy the following relation BAf d (t, T1 , T2 ) = Xf d (t, T2 ) f (T1 , T2 ) d (T1 , T2 ) Pd (t, T1 ) Pd (t, T2 ) . (3.5) Pd (t, T1 ) Xf d (t, T1 ) Pd (t, T2 ) Xf d (t, T1 )

Notice that we could forget the foreign currency analogy above and start by postulating Xf d (t, T ) as in eq. 3.4, name it (discount) basis adjustment and proceed with the next step. We proceed by assuming, according to the standard market practice, the following (driftless) lognormal martingale dynamic for the {f (foreign) forward rate dFf (t; T1 , T2 ) T = f (t) dWf 2 (t) , t T1 , Ff (t; T1 , T2 ) (3.6)

T nian motion Wf 2 under the forwarding (foreign) T2 forward measure QT2 , associated f to the {f (foreign) numeraire Pf (t, T2 ). Next, since Xf d (t, T2 ) in eq. 3.4 is the ratio between the price at time t of a {d (domestic) tradable asset (xf d (t) Pf (t, T2 ) in eq. 3.2, or Pf (t, T2 ) in eq. 3.4 with xf d (t) = 1) and the numeraire Pd (t, T2 ), it must evolve according to a (driftless) martingale process under the associated discounting (domestic) T2 forward measure QT2 , d

where f (t) is the volatility (positive deterministic function of time) of the process, under the probability space , F f , QT2 with the ltration Ftf generated by the browf

dXf d (t, T2 ) T = X (t) dWX2 (t) , t T2 , Xf d (t, T2 )

(3.7)

where X (t) is the volatility (positive deterministic function of time) of the process and T WX2 is a brownian motion under QT2 such that d
T T dWf 2 (t) dWX2 (t) = f X (t) dt.

(3.8)

14

Now, in order to calculate expectations such as in the r.h.s. of eq. 2.13, we must switch from the forwarding (foreign) measure QT2 associated to the numeraire Pf (t, T2 ) f to the discounting (domestic) measure QT2 associated to the numeraire Pd (t, T2 ). In our d double-curve-single-currency language this amounts to transform a cashow on curve {f to the corresponding cashow on curve {d . Recurring to the change-of-numeraire tecnique (cfr. refs. [13], [15], [16]) we obtain that the dynamic of Ff (t; T1 , T2 ) under QT2 acquires a non-zero drift d dFf (t; T1 , T2 ) T = f (t) dt + f (t) dWf 2 (t) , t T1 , Ff (t; T1 , T2 ) f (t) = f (t) X (t) f X (t) , (3.9) (3.10)

and that Ff (T1 ; T1 , T2 ) is lognormally distributed under QT2 with mean and variance d given by Z T1 T Ff (T1 ; T1 , T2 ) 1 2 Qd 2 = ln f (u) f (u) du, (3.11) Et Ff (t; T1 , T2 ) 2 t Z T1 T Ff (T1 ; T1 , T2 ) Qd 2 = ln Vart 2 (u) du. (3.12) f Ff (t; T1 , T2 ) t We thus obtain the following expressions, for t0 t < T1 ,
T Q 2 Et d [Ff (T1 ; T1 , T2 )] = Ff (t; T1 , T2 ) QAf d t, T1 , f , X , f X , R T1 R T1 QAf d t, T1 , f , X , f X = e t f (u)du = e t f (u)X (u)f X (u)du .

(3.13) (3.14)

We conclude that the foreign-currency analogy allows us to compute the expectation of a forward rate on curve {f w.r.t. the discounting measure QT2 associated to the d discounting curve {d through the numeraire Pd (t, T2 ) in terms of a well-known quanto adjustment, typical of cross-currency derivatives. From the discussion above it is clear that such adjustment essentially follows from a change between the probability measures QT2 and QT2 , or numeraires Pf (t, T2 ) and Pd (t, T2 ), naturally associated to the two yield f d curves, {f and {d , respectively. Obviously we can also dene an additive quanto adjustment as
T Q 2 Et d [Ff (T1 ; T1 , T2 )] = Ff (t; T1 , T2 ) + QA0 d t, T1 , f , X , f X , f QA0 d t, T1 , f , X , f X = Ff (t; T1 , T2 ) QAf d t, T1 , f , X , f X 1 , f

(3.15) (3.16)

where the second relation comes from eq. 3.13. Notice that the expressions 3.14 depends on the average over the time interval [t, T1 ] of the product of the volatility f of the 15

{f (foreign) forward rates Ff , of the volatility X of the forward exchange rate Xf d between curves {f and {d , and of the correlation f X between Ff and Xf d . It does not depend either on the volatility d of the {d (domestic) forward rates Fd or on any stochastic quantity after time T1 . The latter fact is actually quite natural, because the stochasticity of the forward rates involved ceases at their xing time T1 . The dependence on the cashow time T2 is actually implicit in eq 3.14, because the volatilities and the correlation involved are exactly those of the forward and exchange rates on the time interval [T1 , T2 ]. Notice in particular that a non-trivial adjustment is obtained if and only if the forward exchange rate Xf d is stochastic ( X 6= 0) and correlated to the forward rate Ff (f X 6= 0); otherwise expression 3.14 collapses to the single curve case QAf d = 1. The volatilities and the correlation in eq. 3.14 can be extracted from market data. In the present market situation, the volatility f can be extracted from quoted cap/oor options on Euribor6M, while for other rate tenors and for X and f X one must resort to historical estimates. Conversely, given a basis adjustment term structure, such that in g. 2.2, we could take f from the market, assume for simplicity f X ' 1 (or any other functional form), and bootstrap out a term structure for the volatility X . Notice that in this way we are also able to compare informations about the internal dynamics of dierent market sub-areas. We will give some numerical estimate of the quanto adjustment in the next section. Finally, we may derive a relation between the quanto and the basis adjustments. Combining eqs. 3.13, 3.15 with eqs. 2.17, 2.18 we obtain BAf d (t, T1 , T2 ) E d [L (T , T )] = t T2 d 1 2 , Q QAf d t, T1 , f , X , f X Et d [Lf (T1 , T2 )] T Q 2 BA0 d (t, T1 , T2 ) QA0 d t, T1 , f , X , f X = Et d [Ld (T1 , T2 ) Lf (T1 , T2 )] f f 3.2. Swap Rates The discussion above can be remapped, with some attention, to swap rates. Given two payment dates vectors T = {T0 , ..., Tn }, S = {S0 , ..., Sm }, T0 = S0 , for the oating and the xed leg of the swap, respectively, the corresponding fair swap rate on curve {f is dened as
n P Q
T2

(3.17) (3.18)

Pf (t, Ti ) f (Ti1 , Ti ) Ff (t; Ti1 , Ti ) Af (t, S) 16 , t T0 , (3.19)

Sf (t, T, S) =

i=1

where Af (t, S) =

is the annuity on curve {f . Following the standard market practice, we observe that, assuming the annuity as the numeraire, the swap rate in eq. 3.19 is the ratio between a tradable asset (the value of the swap oating leg on curve {f ) and the numeraire Af (t, S), and thus it is a martingale under the associated forwarding (foreign) swap measure QS . Hence we can assume, mimicking eq. 3.6, a driftless geometric brownian f motion for the swap rate under QS , f dSf (t, T, S) T,S = f (t, T, S) dWf (t) , t T0 , Sf (t, T, S) (3.21)

m X j=1

Pf (t, Sj ) f (Sj1 , Sj )

(3.20)

where f (t, T, S) is the volatility (positive deterministic function of time) of the process T,S and Wf is a brownian motion under QS . f Then, mimicking the discussion leading to eqs. 3.3-3.4, the following identity
m X j=1

Pd (t, Sj ) d (Sj1 , Sj ) Xf d (t, Sj ) = xf d (t)

= Af (t, S) ,

m X j=1

Pf (t, Sj ) f (Sj1 , Sj ) (3.22)

must hold by no arbitrage between the two curves {f and {d . Dening a swap forward exchange rate Yf d (t, S) such that
m X j=1

Pd (t, Sj ) d (Sj1 , Sj ) Xf d (t, Sj ) = Yf d (t, S)

= Yf d (t, S) Ad (t, S) , Af (t, S) , Ad (t, S)

m X j=1

Pd (t, Sj ) d (Sj1 , Sj ) (3.23)

we obtain the expression Yf d (t, S) = (3.24)

equivalent to eq. 3.4. Hence, since Yf d (t, S) is the ratio between the price at time t of the {d (domestic) tradable asset xf d (t) Af (t, S) and the numeraire Ad (t, S), it must evolve according to a (driftless) martingale process under the associated discounting (domestic) swap measure QS , d dYf d (t, S) S = Y (t, S) dWY (t) , t T0 , Yf d (t, S) 17 (3.25)

where vY (t, S) is the volatility (positive deterministic function of time) of the process S and WY is a brownian motion under QS such that d
T,S S dWf (t) dWY (t) = f Y (t, T, S) dt.

(3.26)

Now, applying again the change-of-numeraire tecnique of sec. 3.1, we obtain that the dynamic of the swap rate Sf (t, T, S) under the discounting (domestic) swap measure QS acquires a non-zero drift d dSf (t, T, S) T,S = f (t, T, S) dt + f (t, T, S) dWf (t) , t T0 , Sf (t, T, S) f (t, T, S) = f (t, T, S) Y (t, S) f Y (t, T, S) , (3.27) (3.28)

and that Sf (t, T, S) is lognormally distributed under QS with mean and variance given d by Z T0 Sf (T0 , T, S) 1 = ln f (u, T, S) 2 (u, T, S) du, Sf (t, T, S) 2 f t Z T0 Sf (T0 , T, S) QS = Vart d ln 2 (u, T, S) du. f Sf (t, T, S) tf Et
QS d

(3.29) (3.30)

We thus obtain the following expressions, for t0 t < T0 , [Sf (T0 , T, S)] = Sf (t, T, S) QAf d t, T, S, f , Y , f Y , (3.31) R T0 R T0 QAf d t, T, S, f , Y , f Y = e t f (u,T,S)du = e t f (u,T,S) Y (u,S)f Y (u,T,S)du . (3.32) Et
QS d

The same considerations as in sec. 3.1 apply. In particular, we observe that the adjustment in eqs. 3.31, 3.33 follows from a change between the probability measures QS and QS , or numeraires Af (t, S) and Ad (t, S), naturally associated to the two yield f d curves, {f and {d , respectively, once swap rates are considered. In the present market situation, the volatility f (u, T, S) in eq. 3.32 can be extracted from quoted swaptions on Euribor6M, while for other rate tenors and for Y (u, S) and f Y (u, T, S) one must resort to historical estimates. An additive quanto adjustment can also be dened as before [Sf (T0 , T, S)] = Sf (t, T, S) + QA0 d t, T, S, f , Y , f Y , f QA0 d t, T, S, f , Y , f Y = Sf (t, T, S) QAf d t, T, S, f , Y , f Y 1 . f Et
QS d

(3.33) (3.34)

18

4. Double-Curve Pricing & Hedging Interest Rate Derivatives

4.1. Pricing The discussion above allows us to derive the no arbitrage, double-curve-single-currency pricing formulas for interest rate derivatives. The recipes are, basically, eqs. 3.13-3.14 or 3.31-3.32. The simplest interest rate derivative is a oating zero coupon bond paying at time T a single cashow depending on a single spot rate (e.g. the Euribor) xed at time t < T, (4.1) ZCB (T ; T, N ) = N f (t, T ) Lf (t, T ) . Being Lf (t, T ) = 1 Pf (t, T ) = Ff (t; t, T ) , f (t, T ) Pf (t, T ) (4.2)

the price at time t T is given by ZCB (t; T, N ) = N Pd (t, T ) f (t, T ) Et

QT d

[Ff (t; t, T )] (4.3)

= N Pd (t, T ) f (t, T ) Lf (t, T ) .

Notice that the basis adjustment in eq. 4.3 disappears and we are left with the standard pricing formula, modied according to the double-curve paradigm. Next we have the FRA, whose payo is given in eq. 2.5 and whose price at time t T1 is given by T Qd 2 FRA (t; T1 , T2 , K, N ) = N Pd (t, T2 ) f (T1 , T2 ) Et [Ff (T1 ; T1 , T2 )] K = N Pd (t, T2 ) f (T1 , T2 ) Ff (t; T1 , T2 ) QAf d t, T1 , f X , f , X K . (4.4)

Notice that in eq. 4.4 for K = 0 and T1 = t we recover the zero coupon bond price in eq. 4.3. For a (payer) oating vs xed swap with payment dates vectors T, S as in sec. 3.2 we have the price at time t T0 Swap (t; T, S, K, N) n X Ni Pd (t, Ti ) f (Ti1 , Ti ) Ff (t; Ti1 , Ti ) QAf d t, Ti1 , f X,i , f,i , X,i =
i=1 m X j=1

Nj Pd (t, Sj ) d (Sj1 , Sj ) Kj .

(4.5)

19

For constant nominal and xed rate the fair (equilibrium) swap rate is given by
n P

Sf (t, T, S) = where

i=1

Pd (t, Ti ) f (Ti1 , Ti ) Ff (t; Ti1 , Ti ) QAf d t, Ti1 , f X,i , f,i , X,i Ad (t, S)
m X j=1

(4.6) Ad (t, S) = Pd (t, Sj ) d (Sj1 , Sj ) (4.7)

is the annuity on curve {d . For caplet/oorlet options on a T1 -spot rate with payo at maturity T2 given by cf (T2 ; T1 , T2 , K, ,N ) = N Max { [Lf (T1 , T2 ) K]} f (T1 , T2 ) ,
T

(4.8)

the standard market-like pricing expression at time t T1 T2 is modied as follows cf (t; T1 , T2 , K, ,N ) = N Et [Max { [Lf (T1 , T2 ) K]} f (T1 , T2 )] = N Pd (t, T2 ) f (T1 , T2 ) Bl Ff (t; T1 , T2 ) QAf d t, T1 , f X , f , X , K, f , f , , (4.9)
Qd 2

where = +/ 1 for caplets/oorlets, respectively, and Bl [F, K, , , ] = F d+ K d , ln + (t, T ) d = (t, T ) Z Z T 2 (u) du, (t, T ) = (t, T ) =
t F K 1 2 2 (t, T ) T

(4.10) (4.11) (4.12)

2 (u) du,

is the standard Black-Scholes formula. Hence cap/oor options prices are given at t T0 by CF (t; T, K, , N) = =
n X i=1 n X i=1

cf (Ti ; Ti1 , Ti , Ki , i ,Ni )

Ni Pd (t, Ti ) f (Ti1 , Ti ) (4.13)

Finally, for swaptions on a T0 -spot swap rate with payo at maturity T0 given by Swaption (T0 ; T, S, K, N ) = NMax [ (Sf (T0 , T, S) K)] Ad (T0 , S) , 20 (4.14)

Bl Ff (t; Ti1 , Ti ) QAf d t, Ti1 , f X,i , f,i , X,i , Ki , f,i , f,i , i ,

the standard market-like pricing expression at time t T0 , using the discounting swap measure QS associated to the numeraire Ad (t, S) on curve {d , is modied as follows d Swaption (t; T, S, K, N ) = N Ad (t, S) Et d {Max [ (Sf (T0 , T, S) K)]} = N Ad (t, S) Bl Sf (t, T, S) QAf d t, T, S, f , Y , f Y , K, f , f , . (4.15) where we have used eq. 3.31 and the quanto adjustment term QAf d t, T, S, f , Y , f Y is given by eq. 3.32. When two or more dierent underlying interest-rates are present, pricing expressions may become more involved. An example is the spread option, for which the reader can refer to, e.g., ch. 14.5.1 in ref. [13]). The calculations above show that also basic interest rate derivatives prices include a quanto adjustment and are thus volatility and correlation dependent. In g. 4.1 we show some numerical scenario for the quanto adjustment in eqs. 3.14, 3.16. We see that, for realistic values of volatilities and correlation, the magnitudo of the additive adjustment may be non negligible, ranging from a few basis points up to over 10 basis points. Time intervals longer than the 6M period used in g. 4.1 further increase the eect. Notice that positive correlation implies negative adjustment, thus lowering the forward rates that enters the pricing formulas above. The standard market practice for pricing interest rate derivatives does not consider the quanto adjustment, thus leaving, in principle, the door open to arbitrage opportunities. In practice the adjustment depends on variables presently not quoted on the market, making virtually impossible to set up arbitrage positions to lock today positive gains in the future. Obviously, it is always possible to bet on a view of the future realizations of the volatilties and correlation. 4.2. Hedging We come now to the problem of hedging. We assume to have a trading portfolio lled with a variety of interest rate derivatives with dierent underlying rate tenors. The rst issue is how to calculate the delta sensitivity. In principle, the answer is straightforward: having recognized interest-rates with dierent tenors as dierent underlyings, and havo n ing constructed multiple yield curves { = {d , {I1 , ..., {IN using homogeneous market f f instruments, we must coherently calculate the delta with respect to the market rate of each bootstrapping instrument8 . In practice this can be computationally cumbersome, given the higher number of market instruments involved. Once the delta sensitivity is known for each pillar of each relevant curve, the next issues of hedging are the choice of the set of hedging instruments and the calculation
8 with the obvious caveat of avoiding double counting of those instruments eventually appearing in more than one curve (3M Futures for instance could appear both in {d and in {I3M curves). f

QS

21

of the corresponding hedge ratios. In principle, there are two alternatives: a) the set of hedging instruments overlaps exactly the set of bootstrapping instruments; or, b) it is a subset restricted to the most liquid bootstrapping instruments. The rst choice allows for a straightforward calculation of hedge ratios and representation of the delta risk distribution of the portfolio. But, in practice, people prefer to hedge using the most liquid instruments, both for better condence in their market prices and for reducing the cost of hedging. Hence the second strategy generally prevails. In this case the calculation of hedge ratios requires a three-step procedure: rst, the delta is calculated on the basis of all bootstrapping instruments; second, it is re-aggregated, pillar by pillar, on the basis of hedging instruments, using the appropriate mapping rules; then, hedge ratios are calculated. The disadvantage of this second choice is, clearly, that some risk the basis risk in particular - is only partially hedged: hence, a particular care is required in the choice of the hedging instruments. A nal issue regards portfolio management. In principle one could keep all the interest rate derivatives together in a single portfolio, pricing each one with its appropriate forwarding curve, discounting all cashows with the same discounting curve, and hedging using the preferred choice described above. A possible alternative is the segregation of homogeneous contracts (with the same underlying interest rate index) into dedicated sub-portfolios, each managed with its appropriate curves and hedging tecniques. The eventually remaining non-homogeneous instruments (those not separable in pieces depending on a single underlying) can be redistributed in the portfolios above according to their prevailing underlying (if any), or put in other isolated portfolios, to be handled with special care. The main advantage of this second approach is to clean up the trading books, cornering the more complex deals in a controlled way, and to allow a clearer and self-consistent representation of the sensitivities to the dierent underlyings, and in particular of the basis risk of each sub-portfolio, thus allowing for a cleaner hedging.

5. Conclusions
We have discussed how the liquidity crisis and the resulting changes in the market quotations, in particular the very high basis swap spreads, have forced the market practice to evolve the standard procedure adopted for pricing and hedging single-currency interest rate derivatives. The new double-curve paradigm involves the bootstrapping of multiple yield curves using separated sets of vanilla interest rate instruments homogeneous in the underlying rate (typically with 1M, 3M, 6M, 12M tenors). Prices, sensitivities and hedge ratios of interest rate derivatives on a given underlying rate tenor are calculated using the corresponding forward curve with the same tenor, plus a second distinct curve for discount factors. We have shown that the old, well-known, standard single-curve no arbitrage relations 22

are no longer valid and can be recovered with the introduction of a (forward) basis adjustment, for which simple statical expressions are given in eqs. 2.17-2.18 in terms of discount factors from the two curves. Our numerical results have shown that the basis adjustment curves, in particular in a realistic stressed market situation, may display an oscillating term structure, not present in the smooth and monotonic basis swaps market quotes and more complex than that of the discount and forward curves. Such richer micro-term structure is caused by amplication eects of small local dierences between the discount and forwarding curves and constitutes both a very sensitive test of the quality of the bootstrapping procedure (interpolation in particular), and an indicator of the tiny, but observable, dierences between dierent interest rate market areas. Both of these causes may have appreciable eects on the price of interest rate instruments, in particular when one switches from the single-curve towards the double-curve framework. Recurring to the foreign-currency analogy we have also been able to recompute the no arbitrage double-curve-single-currency market-like pricing formulas for basic interest rate derivatives, zero coupon bonds, FRA, swaps caps/oors and swaptions in particular. Such prices depend on forward or swap rates on curve {f corrected with the wellknown quanto adjustment typical of cross-currency derivatives, naturally arising from the change between the numeraires, or probability measures, naturally associated to the two yield curves. The quanto adjustment depends on the volatility f of the forward rates Ff on {f , of the volatility X of the forward exchange rate Xf d between {f and {d , and of the correlation f X between Ff and Xf d . In particular, a non-trivial adjustment is obtained if and only if the forward exchange rates Xf d are stochastic ( X 6= 0) and correlated to the forward rate Ff (f X 6= 0). Analogous considerations hold for the swap rate quanto adjustment. Numerical scenarios show that the quanto adjustment can be non negligible for realistic values of volatilities and correlation. The standard market practice does not take into account the adjustment, thus being, in principle, not arbitrage free, but, in practice the market does not trade enough instruments to set up arbitrage positions. Hence only bets are possible on the future realizations of voloatility and correlation.

References
[1] P. S. Hagan, G. West, Methods for Constructing a Yield Curve, Wilmott Magazine, p 70-81, May 2008. [2] P. S. Hagan, G. West, Interpolation Methods for Curve Construction, Applied Mathematical Finance, Vol. 13, No. 2, 89129, June 2006. [3] L. Andersen, Discount Curve Construction with Tension Splines, Review of Derivatives Research, Springer, vol. 10, issue 3, pages 227-267, Dec. 2007. 23

[4] U. Ron, A Practical Guide to Swap Curve Construction, Working Paper 2000-17, Bank of Canada, Aug. 2000. [5] P. Madigan, Libor Under Attack, Risk Magazine, http://www.risk.net/public/showPage.html?page=797090. Jun. 2008,

[6] F. Ametrano, M. Bianchetti, Smooth Yield Curves Bootstrapping For Forward Libor Rate Estimation and Pricing Interest Rate Derivatives, to be published in Modelling Interest Rates: Latest Advances for Derivatives Pricing, edited by F. Mercurio, Risk Books. [7] E. Fruchard, C. Zammouri, E. Willems, Basis for change, Risk, 8(10), 7075, 1995. [8] B. Tuckman, P. Porrio, Interest Rate Parity, Money Market Basis Swaps, and Cross-Currency Basis Swaps, Fixed Income Liquid Markets Research, Lehman Brothers, Jun 2003. [9] W. Boenkost, W. Schmidt, Cross currency swap valuation, Working Paper, HfB Business School of Finance & Management, May 6, 2005, http://www.frankfurtschool.de/dms/publications-cqf/CPQF_Arbeits2.pdf. [10] M. Kijima, K. Tanaka, T. Wong, A Multi-Quality Model of Interest Rates, Quantitative Finance, 2008. [11] M. Morini, Credit Modelling After the Subprime Crisis, Marcus Evans course, 2008. [12] F. Mercurio, Post Credit Crunch Interest Rates: Formulas and Market Models, working paper, Bloomberg, 2008, http://ssrn.com/abstract=1332205. [13] D. Brigo, F. Mercurio, Interest Rate Models - Theory and Practice, 2nd edition, Springer 2006. [14] QuantLib is an open-source (http://www.quantilib.org). object oriented C++ nancial library

[15] F. Jamshidian, An Exact Bond Option Formula, Journal of Finance, 44, pp. 205-209, 1989. [16] H. Geman, N. El Karoui, J.C.Rochet, Changes of Numeraire, Changes of Probability Measure and Option Pricing, J. of Applied Probability, Vol. 32, n. 2, pp. 443-458, 1995.

24

Figure 1.1: Quotations as of 30 Sep. 2008 for ve basis swap curves corresponding to the four Euribor swap curves 1M, 3M, 6M, 12M (source: Reuters, contributor: ICAP).

25

110 100 90 80

EUR Basis swaps

basis spread (bps)

3M 1M 1M 6M 3M 1M

vs vs vs vs vs vs

6M 3M 6M 12M 12M 12M

70 60 50 40 30 20 10 0 1Y 2Y 3Y 4Y 5Y 6Y 7Y 8Y 9Y 10Y 11Y 12Y 15Y 20Y 25Y 30Y

Figure 1.2: Euribor basis swap spreads (basis points) versus swap maturity (years) from Fig. 1.1. The curve 1M-12M has been deduced from the quoted 3M-12M and 1M-3M curves.

26

5.00% 4.75% 4.50% forward rate (%) 4.25% 4.00% 3.75% 3.50% 3.25% 3.00% 09/2008 09/2011 09/2014 09/2017 09/2020 09/2023 09/2026 09/2029 09/2032

6M curve

09/2035

09/2038

09/2041

09/2044

09/2047

09/2050

09/2053

09/2056

09/2059

09/2062 09/2062

5.00% 4.75% 4.50% forward rate (%) 4.25% 4.00% 3.75% 3.50% 3.25% 3.00% 09/2008 09/2011 09/2014 09/2017 09/2020 09/2023 09/2026 09/2029

Discount curve

09/2032

09/2035

09/2038

09/2041

09/2044

09/2047

09/2050

09/2053

09/2056

09/2059

Figure 2.1: Forward curves, plotted with 6M-tenor forward rates F (t0 ; t, t + 6M, act/360 ), t daily sampled and t0 = 15 Sep. 2008. Upper panel: forward curve from 6M curve {I6M ; lower panel: forward curve from discount curve {d f (see description in the text). The eects of the market stress are clearly visible in the crazy roller-coaster look-up of the curves. The same pattern is observed also in the 1M, 3M, 12M curves (not reported here).

27

09/2065

09/2065

1.15 1.10 1.05

Basis adjustment (multiplicative)

1.03 1.02 basis adjustment 1.01 1.00 0.99 0.98 0.97

Basis adjustment (multiplicative)

basis adjustment

1.00 0.95 0.90 0.85 0.80 0.75 Mar-09 Mar-10 Dec-08 Dec-09 Dec-10 Mar-11 Jun-09 Jun-10 Sep-08 Jun-11 Sep-09 Sep-10 Sep-11 1M vs Disc 3M vs Disc 6M vs Disc 12M vs Disc

1M vs Disc 3M vs Disc 6M vs Disc 12M vs Disc Sep-11 Sep-14 Sep-17 Sep-20 Sep-23 Sep-26 Sep-29 Sep-32 Sep-35 Sep-38 Sep-41 Sep-44 Sep-44 Sep-47 Sep-47

120 100

Basis adjustment (additive)

15 10

Basis adjustment (additive)

1M vs Disc 3M vs Disc 6M vs Disc 12M vs Disc

80 basis spread (bps) 60 40 20 0 -20

basis spread (bps)

1M vs Disc 3M vs Disc 6M vs Disc 12M vs Disc

5 0 -5

-10 -40 -60 Mar-09 Mar-10 Mar-11 Jun-09 Jun-10 Dec-08 Dec-09 Dec-10 Sep-08 Sep-09 Sep-10 Jun-11 Sep-11 -15 Sep-11 Sep-14 Sep-17 Sep-20 Sep-23 Sep-26 Sep-29 Sep-32 Sep-35 Sep-38 Sep-41

Figure 2.2: Upper panels: multiplicative basis adjustments from eq. 2.17 as of 15 Sep. 2008 (end of day), for daily sampled 6M-tenor forward rates as in g. 2.1, calculated on {I1M , {I3M , {I6M and {I12M curves against {d taken as reference curve. Lower panels: f f f f equivalent plots of the additive basis adjustment of eq. 2.18 between the same forward rates (basis points). Left panels: 0Y-3Y data; Right panels: 3Y-40Y data on magnied scales. The higher short-term adjustments seen in the left panels are due to the higher short-term market basis spread (see Figs. 1.1-1.2). The oscillating term structure observed is due to the amplication of small dierences in the term structures of the curves.

28

25 20 15 basis spread (bps) 10 5 0 -5

Basis adjustment (additive)

1M vs Disc 3M vs Disc 6M vs Disc 12M vs Disc

-10 -15 -20 -25 Sep-08 Sep-10 Sep-12 Sep-14 Sep-16 Sep-18 Sep-20 Sep-22 Sep-24 Sep-26 Sep-28 Sep-30 Sep-32 Sep-34 Sep-36 Sep-38 Sep-40 Sep-42 Sep-44 Sep-46 09/2065 Sep-48

5.00% 4.75% 4.50% forward rate (%) 4.25% 4.00% 3.75% 3.50% 3.25% 3.00% 09/2008 09/2011 09/2014 09/2017 09/2020 09/2023 09/2026

Forward curve 6M

09/2029

09/2032

09/2035

09/2038

09/2041

09/2044

09/2047

09/2050

09/2053

09/2056

09/2059

Figure 2.3: Same as in gs. 2.1 and 2.2, but with linear interpolation on zero rates (a common market practice). The angular points in the lower panel clearly show the inadequacy of the boostrap, but the very strong oscillations in the (additive) basis adjustment in the upper panel (notice the dierent scales w.r.t. g. 2.2, lower panels) allows to further appreciate the innatural dierences induced in similar forward instruments priced on the two curves.

29

09/2062

Figure 3.1: Picture of no-arbitrage interpretation for the forward exchange rate in eq. 3.2. Moving, in the yield curve vs time plane, from top right to bottom left corner through path A or path B must be equivalent. Alternatively, we may think to noarbitrage as a sort of zero circuitation, sum of all trading events following a closed path starting and stopping at the same point in the plane. This description is equivalent to the traditional table of transaction picture, as found e.g. in g. 1 of ref. [8].

30

1.04 1.03 1.02 Quanto adj. 1.01 1.00 0.99 0.98 0.97 0.96 -1.0 -0.8

Quanto Adjustment (multiplicative)

Sigma_f = 20%, Sigma_X = 5% Sigma_f = 30%, Sigma_X = 10% Sigma_f = 40%, Sigma_X = 20% -0.6 -0.4 -0.2 -0.0 0.2 Correlation 0.4 0.6 0.8 1.0

20 15 Quanto adj. (bps) 10 5 0 -5 -10 -15 -20 -1.0 -0.8 -0.6

Quanto Adjustment (additive)

Sigma_f = 20%, Sigma_X = 5% Sigma_f = 30%, Sigma_X = 10% Sigma_f = 40%, Sigma_X = 20% -0.4 -0.2 -0.0 0.2 Correlation 0.4 0.6 0.8 1.0

Figure 4.1: Numerical scenarios for the quanto adjustment. Upper panel: multiplicative (from eq. 3.14); lower panel: additive (from eq. 3.16). In each gure we show the quanto adjustment corresponding to three dierent combinations of (at) volatility values as a function of the correlation. The time interval is xed to T1 t = 0.5 and the forward rate entering eq. 3.16 to 4%, a typical value in g. 2.1. We see that, for realistic values of volatilities and correlation, the magnitudo of the additive adjustment may be important. 31