CONVEXITY CONUNDRUMS: PRICING CMS SWAPS, CAPS, AND FLOORS

PATRICK S. HAGAN
GORILLA SCIENCE
11 PALISADE PLAZA
EDGEWATER, NJ 07020
PHAGAN1954@HOTMAIL.COM

1. Introduction. Here we focus on a single class of deals, the constant maturity swaps, caps, and
floors. We develop a framework that leads to the standard methodology for pricing these deals, and then
use this framework to systematically improve the pricing.
Let us start by agreeing on basic notation. In our notation, today is always w = 0. We use

(1.1a) ](w; W ) = value at date w of a zero coupon bond with maturity W>
(1.1b) G(W )  ](0> W ) = today’s discount factor for maturity W=

We distinguish between zero coupon bonds and discount factors to remind ourselves that discount factors
are not random, we can always obtain the current discount factors G(W ) by stripping the yield curve, while
zero coupon bonds ](w> W ) remain random until the present catches up to date w= We also use

(1.2) cvg(wvw > whqg > dcb)

to denote the coverage (also called the year fraction or day count fraction) of the period wvw to whqg , where
dcb is the day count basis (Act360, 30360, . . . ) specified by the contract. So if interest accrues at rate U,
then cvg(wvw > whqg > dcb)U is the interest accruing in the interval wvw to whqg .
1.1. Deal definition. Consider a CMS swap leg paying, say, the Q year swap rate plus a margin p.
Let w0 > w1 > = = = > wp be the dates of the CMS leg specified in the contract. (These dates are usually quarterly).
For each period m, the CMS leg pays

(1.3a)  m (Um + p) paid at wm for m = 1> 2> = = = > p>

where Um is the Q year swap rate and

(1.3b)  m = cvg(wm1 > wm > dcbsd| )

is the coverage of interval m. If the CMS leg is set-in-advance (this is standard), then Um is the rate for
a standard swap that begins at wm1 and ends Q years later. This swap rate is fixed on the date  m that
is spot lag business days before the interval begins at wm1 , and pertains throughout the interval, with the
accrued interest  m (Um + p) being paid on the interval’s end date, wm . Although set-in-advance is the market
standard, it is not uncommon for contracts to specify CMS legs set-in-arrears. Then Um is the Q year swap
rate for the swap that begins on the end date wm of the interval, not the start date, and the fixing date  m for
Um is spot lag business days before the interval ends at wm . As before,  m is the coverage for the m wk interval
using the day count basis dcbsd| specified in the contract. Standard practice is to use the 30360 basis for
USD CMS legs.
CMS caps and floors are constructed in an almost identical fashion. For CMS caps and floosr on the Q
year swap rate, the payments are

(1.4a)  m [Um  N]+ paid at wm for m = 1> 2> = = = > p> (cap),
(1.4b)  m [N  Um ]+ paid at wm for m = 1> 2> = = = > p> (floor),

where the Q year swap rate is set-in-advance or set-in-arrears, as specified in the contract.
1
 1.2. Reference swap. The value of the CMS swap, cap, or floor is just the sum of the values of each
payment. Any margin payments p can also be valued easily. So all we need do is value a single payment of
the three types,
(1.5a) Uv paid at ws >
(1.5b) [Uv  N]+ paid at ws >
(1.5c) [N  Uv ]+ paid at ws =
Here the reference rate Uv is the par rate for a standard swap that starts at date v0 , and ends Q years
later at vq . To express this rate mathematically, let v1 > v2 > = = = > vq be the swap’s (fixed leg) pay dates. Then
a swap with rate Ui l{ has the fixed leg payments
(1.6a) m Ui l{ paid at vm for m = 1> 2> = = = > q>
where
(1.6b) m = cvg(wm1 > wm > dcbvz )
is the coverage (fraction of a year) for each period m, and dcbvz is the standard swap basis. In return
for making these payments, the payer receives the floating leg payments. Neglecting any basis spread, the
floating leg is worth 1 paid at the start date v0 , minus 1 paid at the end date vq . At any date w, then, the
value of the swap to the payer is
q
X
(1.7) Yvz (w) = ](w; v0 )  ](w; vq )  Ui l{ m ](w; vm )=
m=1

The level of the swap (also called the annuity, PV01, DV01, or numerical duration) is defined as
q
X
(1.8) O(w) = m ](w; vm )=
m=1

Crudely speaking, the level O(w) represents the value at time w of receiving $1 per year (paid annually or
semiannually, according to the swap’s frequency) for Q years. With this definition, the value of the swap is
(1.9a) Yvz (w) = [Uv (w)  Ui l{ ]O(w)>
where
](w; v0 )  ](w; vq )
(1.9b) Uv (w) = =
O(w)
Clearly the swap is worth zero when Ui l{ equals Uv (w), so Uv (w) is the par swap rate at date w. In particular,
today’s level
q
X q
X
(1.10a) O0 = O(0) = m Gm = m G(vm )>
m=1 m=1

and today’s (forward) swap rate

G0  Gq
(1.10b) Uv0 = Uv (0) =
O0
are both determined by today’s discount factors.
2
 2. Valuation. According to the theory of arbitrage free pricing, we can choose any freely tradeable
instrument as our numeraire. Examining 1.8 shows that the level O(w) is just the value of a collection zero
coupon bonds, since the coverages m are just fixed numbers. These are clearly freely tradeable instruments,
so we can choose the level O(w) as our numeraire.1 The usual theorems then guarantee that there exists
a probability measure such that the value Y (w) of any freely tradeable deal divided by the numeraire is a
Martingale. So
½ ¯ ¾
Y (W ) ¯¯
(2.1) Y (w) = O(w) E Fw for any W A w>
O(W ) ¯

provided there are no cash flows between w and W .

It is helpful to examine the valuation of a plain vanilla swaption. Consider a standard European option
on the reference swap. The exercise date of such an option is the swap’s fixing date  , which is spot-lag
business days before the start date v0 . At this exercise date, the payo is the value of the swap, provided
this value is positive, so

(2.2) Yrsw ( ) = [Uv ( )  Ui l{ ]+ O( )

on date  . Since the Martingale formula 2.1 holds for any W A w, we can evaluate it at W =  , obtaining
½ ¯ ¾
Yrsw ( ) ¯¯ © ¯ ª
(2.3) Yrsw (w) = O(w) E ¯ Fw = O(w) E [Uv ( )  Ui l{ ]+ ¯ Fw =
O( )

In particular, today’s value of the swaption is

© ¯ ª
(2.4a) Yrsw (w) = O0 E [Uv ( )  Ui l{ ]+ ¯ F0 =

Moreover, 1.9b shows that the par swap rate Uv (w) is the value of a freely tradable instrument (two zero
coupon bonds) divided by our numeraire. So the swap rate must also a Martingale, and

(2.4b) E { Uv ( )| F0 } = Uv (0)  Uv0 =

To complete the pricing, one now has to invoke a mathematical model (Black’s model, Heston’s model,
the SABR model, . . . ) for how Uv ( ) is distributed around its mean value Uv0 . In Black’s model, for example,
the swap rate is distributed according to
s
  12  2 
(2.5a) Uv ( ) = Uv0 h{ >

where { is a normal variable with mean zero and unit variance. In the “normal” or “absolute vol”model, the
swap rate is distributed according to
s
(2.5b) Uv ( ) = Uv0 + d{  >

where again { is a normal variable with mean zero and unit variance. One completes the pricing by integrating
to calculate the expected value.

1 We follow the standard (if bad) practice of referring to both the physical instrument and its value as the “numeraire.”
3
 2.1. CMS caplets. The payo of a CMS caplet is

(2.6) [Uv ( )  N]+ paid at ws =

On the swap’s fixing date  , the par swap rate Uv is set and the payo is known to be [Uv ( )  N]+ ]( ; ws ),
since the payment is made on ws . Evaluating 2.1 at W =  yields
½ ¯ ¾
FPV [Uv ( )  N]+ ]( ; ws ) ¯¯
(2.7a) Yfds (w) = O(w) E ¯ Fw =
O( )
In particular, today’s value is
½ ¯ ¾
FPV [Uv ( )  N]+ ]( ; ws ) ¯¯
(2.7b) Yfds (0) = O0 E ¯ F0 =
O( )
The ratio ]( ; ws )@O( ) is (yet another!) Martingale, so it’s average value is today’s value:

(2.8) E { ]( ; ws )@O( )| F0 } = G(ws )@O0 =

By dividing ]( ; ws )@O( ) by its mean, we obtain
½ ¯ ¾
FPV
¯
+ ]( ; ws )@O( ) ¯
(2.9) Yfds (0) = G(ws ) E [Uv ( )  N] F0 >
G(ws )@O0 ¯
which can be written more evocatively as
© ¯ ª
(2.10) FPV
Yfds (0) = G(ws ) E [Uv ( )  N]+ ¯ F0
½ µ ¶¯ ¾
]( ; ws )@O( ) ¯
+G(ws ) E [Uv ( )  N]+  1 ¯¯ F0 =
G(ws )@O0

The first term is exactly the price of a European swaption with notional G(ws )@O0 , regardless of how
the swap rate Uv ( ) is modeled. The last term is the “convexity correction.” Since Uv ( ) is a Martingale
and []( ; ws )@O( )] @ [](w; ws )@O(w)]  1 is zero on average, this term goes to zero linearly with the variance
of the swap rate Uv ( ), and is much, much smaller than the first term.
There are two steps in evaluating the convexity correction. The first step is to model the yield curve
movements in a way that allows us to re-write the level O( ) and the zero coupon bond ]( ; ws ) in terms of
the swap rate Uv . (One obvious model is to allow only parallel shifts of the yield curve). Then we can write

(2.11a) ]( ; ws )@O( ) = J(Uv ( ))>

(2.11b) G(ws )@O0 = J(Uv0 )>

for some function J(Uv ). The convexity correction is then just the expected value
½ µ ¶¯ ¾
J(Uv ( )) ¯
(2.12) ff = G(ws ) E [Uv ( )  N] +
 1 ¯¯ F0
J(Uv0 )
over the swap rate Uv ( ). The second step is to evaluate this expected value.
In the appendix we start with the street-standard model for expressing O( ) and ]( ; ws ) in terms of
the swap rate Uv . This model uses bond math to obtain
Uv 1
(2.13a) J(Uv ) = 1 =
(1 + Uv @t) 1  (1+Uv @t)q
4
Here t is the number of periods per year (1 if the reference swap is annual, 2 if it is semi-annual, ...), and
ws  v0
(2.13b) =
v1  v0
is the fraction of a period between the swap’s start date v0 and the pay date ws . For deals “set-in-arrears”
 = 0. For deals “set-in-advance,” if the CMS leg dates w0 > w1 > = = = are quarterly, then ws is 3 months after
the start date v0 , so  = 12 if the swap is semiannual and  = 14 if it is annual.
In the apprendix we also consider sophisticated models for expressing O( ) and ]( ; ws ) in terms of the
swap rate Uv , and obtain increasingly sophisticated functions J(Uv ).
We can carry out the second step by replicating the payo in 2.12 in terms of payer swaptions. For any
smooth function i (Uv ) with i (N) = 0, we can write
Z 4 ½
0 + + 00 i (Uv ) for Uv A N
(2.14) i (N)[Uv  N] + [Uv  {] i ({)g{ = =
N 0 for Uv ? N

Choosing
µ ¶
J({)
(2.15) i ({)  [{  N] 1 >
J(Uv0 )
and substituting this into 2.12, we find that
½ Z 4 ¾
© ¯ ª © ¯ ª
(2.16) ff = G(ws ) i 0 (N)H [Uv ( )  N]+ ¯ F0 + i 00 ({)H [Uv ( )  {]+ ¯ F0 g{ =
N

Together with the first term, this yields

½ Z 4 ¾
FPV G(ws ) 0 00
(2.17a) Yfds (0) = [1 + i (N)] F(N) + F({)i ({)g{ >
O0 N

as the value of the CMS caplet, where

© ¯ ª
(2.17b) F({) = O0 E [Uv ( )  {]+ ¯ F0

is the value of an ordinary payer swaption with strike {.

This formula replicates the value of the CMS caplet in terms of European swaptions at dierent strikes
{. At this point some pricing systems break the integral up into 10bp or so buckets, and re-write the
convexity correction as the sum of European swaptions centered in each bucket. These swaptions are then
consolidated with the other European swaptions in the vanilla book, and priced in the vanilla pricing system.
This “replication method” is the most accurate method of evaluating CMS legs. It also has the advantage
of automatically making the CMS pricing and hedging consistent with the desk’s handling of the rest of its
vanilla book. In particular, it incorporates the desk’s smile/skew corrections into the CMS pricing. However,
this method is opaque and compute intensive. After briefly considering CMS floorlets and CMS swaplets,
we develop simpler approximate formulas for the convexity correction, as an alternative to the replication
method.
2.2. CMS floorlets and swaplets. Repeating the above arguments shows that the value of a CMS
floorlet is given by
( Z N )
G(ws )
(2.18a) YiFPV
orru (0) = [1 + i 0 (N)] S (N)  S ({)i 00 ({)g{ >
O0 4
5
where i ({) is the same function as before (see 2.15), and where
© ¯ ª
(2.18b) S ({) = O0 E [{  Uv ( )]+ ¯ F0

is the value of the ordinary receiver swaption with strike {. Thus, the CMS floolets can also be priced
through replication with vanilla receivers. Similarly, the value of a single CMS swap payment is
(Z Z U0v )
4
FPV G(w s )
(2.19a) Yvzds (0) = G(ws )Uv0 + 00
F({)idwp ({)g{ + 00
S ({)idwp ({)g{ >
O0 Uv0 4

where
µ ¶
J({)
(2.19b) idwp ({)  [{  Uv0 ] 1
J(Uv0 )

is the same as i ({) with the strike N replaced by the par swap rate Uv0 . Here, the first term in 2.19a is
the value if the payment were exactly equal to the forward swap rate Uv0 as seen today. The other terms
represent the convexity correction, written in terms of vanilla payer and receiver swaptions. These too can
be evaluated by replication.
It should be noted that CMS caplets and floorlets satisfy call-put parity. Since

(2.20) [Uv ( )  N]+  [N  Uv ( )]+ = Uv ( )  N paid at ws >

the payo of a CMS caplet minus a CMS floorlet is equal to the payo of a CMS swaplet minus N. Therefore,
the value of this combination must be equal at all earlier times as well:
FPV
(2.21a) Yfds (w)  YiFPV FPV
orru (w) = Yvzds (w)  N](w; ws )

In particular,
FPV
(2.21b) Yfds (0)  YiFPV FPV
orru (0) = Yvzds (0)  NG(ws )=

Accordingly, we can price an in-the-money caplet or floorlet as a swaplet plus an out-of-the-money floorlet
or caplet.
3. Analytical formulas. The function J({) is smooth and slowly varying, regardless of the model
used to obtain it. Since the probable swap rates Uv ( ) are heavily concentrated around Uv0 , it makes sense
to expand J({) as

(3.1a) J({)  J(Uv0 ) + J0 (Uv0 )({  Uv0 ) + · · · =

For the moment, let us limit the expansion to the linear term. This makes i ({) a quadratic function,

J0 (Uv0 )
(3.1b) i ({)  ({  Uv0 )({  N)>
J(Uv0 )
and i 00 ({) a constant. Substituting this into our formula for a CMS caplet (2.17a), we obtain
½ Z 4 ¾
FP V G(ws ) 0 0 0
(3.2) Yfds (0) = F(N) + J (Uv ) (N  Uv )F(N) + 2 F({)g{ >
O0 N

where we have used J(Uv0 ) = G(ws )@O0 . Now, for any N the value of the payer swaption is
© ¯ ª
(3.3a) F(N) = O0 E [Uv ( )  N]+ ¯ F0 >
6
so the integral can be re-written as
Z 4 ½Z 4 ¯ ¾
¯
(3.3b) F({)g{ = O0 E [Uv ( )  {] g{¯¯ F0
+
N
n ¡N ¢2 ¯¯ o
= 2 O0 E [Uv ( )  N]+ ¯ F0 =
1

Putting this together yields

G(ws ) ©£ ¤ ¯ ª
(3.4a) FPV
Yfds (0) = F(N) + J0 (Uv0 )O0 E Uv ( )  Uv0 [Uv ( )  N]+ ¯ F0
O0
for the value of a CMS caplet, where the convexity correction is now the expected value of a quadratic
“payo.” An identical arguments yields the formula

G(ws ) ©£ ¤ ¯ ª
(3.4b) YiFPV
orru (0) = S (N)  J0 (Uv0 )O0 E Uv0  Uv ( ) [N  Uv ( )]+ ¯ F0
O0
for the value of a CMS floorlet. Similarly, the value of a CMS swap payment works out to be

FPV
n¡ ¢2 ¯¯ o
(3.4c) Yvzds (0) = G(ws )Uv0 + J0 (Uv0 )O0 E Uv ( )  Uv0 ¯ F0 =

To finish the calculation, one needs an explicit model for the swap rate Uv ( ). There are two simple
models one can use. The first is Black’s model, which assumes the swap rate Uv ( ) is log normal with a
volatility ,

(3.5a) gUv = Uv gZ>

With this model, one obtains

FPV
¡ ¢2 h 2  i
(3.5b) Yvzds (0) = G(ws )Uv0 + J0 (Uv0 )O0 Uv0 h 1

for the CMS swaplets,

G(ws ) h 2
i
FPV
(3.5c) Yfds (0) = F(N) + J0 (Uv0 )O0 (Uv0 )2 h  N (g3@2 )  Uv0 (Uv0 + N)N (g1@2 ) + Uv0 NN (g1@2 )
O0
for CMS caplets, and
(3.5d)
G(ws ) h 2
i
YiFPV
orru (0) = S (N)  J0 (Uv0 )O0 (Uv0 )2 h  N (g3@2 )  Uv0 (Uv0 + N)N (g1@2 ) + Uv0 NN (g1@2 )
O0
for CMS floorlets. Here

ln Uv0 @N +  2 
(3.5e) g = s =
 

The second model is the normal, or absolute model, which assumes that the swap rate follows

(3.6a) gUv = dgZ>

7
where d is the “absolute” or “normal” vol. This yields
FPV
(3.6b) Yvzds (0) = G(ws )Uv0 + J0 (Uv0 )O0 d2 
for the CMS swaplets,
µ ¶
FPV G(ws ) Uv0  N
(3.6c) Yfds (0) = F(N) + J0 (Uv0 )O0 d2  N s
O0 d 
for CMS caplets, and
µ ¶
G(ws ) N  Uv0
(3.6d) YiFPV
orru (0) = S (N)  J0 (Uv0 )O0 d2  N s
O0 d 
for CMS floorlets. We can obtain the normal vol d by noting that if  is the log normal volatility for a
swaption with forward rate Uv0 and strike N, then the normal volatility d of this swaption is
Uv0  N 1
(3.7a) d= 1 2 1
log Uv0 @N 1 + 24   + 4 2
5760  
¡ ¢
Near the money ( Uv0  N @N is less than 20% or so, we can replace this formula with
p 1+ 1
24 log2 Uv0 @N + 1920
1
log4 Uv0 @N
(3.7b) d= Uv0 N 1 2 1
1 + 24   + 5760 4  2
The key concern with Black’s model is that it does not address the smiles and/or skews seen in the
marketplace. This can be partially mitigated by using the correct volatilities. For CMS swaps, the volatility
 DW P for at-the-money swaptions should be used, since the expected value 3.4c includes high and low strike
swaptions equally. For out-of-the-money caplets and floorlets, the volatility N for strike N should be used,
since the swap rates Uv ( ) near N provide the largest contribution to the expected value. For in-the-money
options, the largest contributions come from swap rates Uv ( ) near the mean value Uv0 . Accordingly, call-
put parity should be used to evaluate in-the-money caplets and floorlets as a CMS swap payment plus an
out-of-the-money floorlet or caplet.
4. Conclusions. The standard pricing for CMS legs is given by 3.5b - 3.5e with J(Uv ) given by 2.13a.
These formula are adequate for many purposes. When finer pricing is required, one can systematically
improve these formulas by using the more sophisticated models for J(Uv ) developed in the Appendix, and
by adding the quadratic and higher order terms in the expansion 3.1a. In addition, 3.4a - 3.4b show that
the convexity corrections are essentially swaptions with “quadratic” payos. These payos emphasize away-
from-the-money rates more than standard swaptions, so the convexity corrections can be quite sensitive to
the market’s skew and smile. CMS pricing can be improved by replacing Black’s model with a model that
matches the market smile, such as Heston’s model or the SABR model. Alternatively, when the very highest
accuracy is needed, replication can be used to obtain near perfect results.
Appendix A. Models of the yield curve.
A.1. Model 1: Standard model. The standard method for computing convexity corrections uses
bond math approximations: payments are discounted at a flat rate, and the coverage (day count fraction)
for each period is assumed to be 1@t, where t is the number of periods per year (1 for annual, 2 for semi-
annual, etc). At any date w, the level is approximated as
q
X Xq
](w> vm ) 1@t
(A.1) O(w) = ](w> v0 ) m  ](w> v0 ) m
>
m=1
](w> v0 ) m=1
[1 + Uv (w)@t]

8
which works out to
· ¸
](w> v0 ) 1
(A.2a) O(w) = 1 =
Uv (w) (1 + Uv (w)@t)q
Here the par swap rate Uv (w) is used as the discount rate, since it represents the average rate over the life of
the reference swap. In a similar spirit, the zero coupon bond for the pay date ws is approximated as
](w> v0 )
(A.2b) ](w; ws )  
>
(1 + Uv (w)@t)
where
ws  v0
(A.2c) =
v1  v0
is the fraction of a period between the swap’s start date v0 and the pay date ws . Thus the standard “bond
math model” leads to
](w; ws ) Uv 1
(A.3) J(Uv ) =  1 =
O(w) (1 + Uv @t) 1  (1+Uv @t)q

This method a) approximates the schedule and coverages for the reference swaption; b) assumes that
the initial and final yield curves are flat, at least over the tenor of the reference swaption; and c) assumes a
correlation of 100% between rates of diering maturities.
A.2. Model 2: “Exact yield” model. We can account for the reference swaption’s schedule and day
count exactly by approximating
m
Y 1
(A.4) ](w; vm )  ](w; v0 ) >
1 + n Uv (w)
n=1

wk
where n is the coverage of the n period of the reference swaption. At any date w, the level is then
q q
à m !
X X Y 1
(A.5) O(w) = m ](w; vm ) = ](w; v0 ) m =
m=1 m=1
1 + n Uv (w)
n=1

We can establish the following identity by induction:

à q
!
](w; v0 ) Y 1
(A.6) O(w) = 1 =
Uv (w) [1 + n Uv (w)]
n=1

In the same spirit, we can approximate

1
(A.7) ](w; ws ) = ](w; v0 ) >
(1 + 1 Uv (w))
where  = (ws  v0 )@(v1  v0 ) as before=Then
](w; ws ) Uv 1
(A.8) J(Uv ) =  q =
O(w) (1 + 1 Uv ) Y
1
1 (1+m Uv )
n=1

This approximates the yield curve as flat and only allows parallel shifts, but has the schedule right.
9
 A.3. Model 3: Parallel shifts. This model takes into account the initial yield curve shape, which
can be significant in steep yield curve environments. We still only allow parallel yield curve shifts, so we are
approximate
](w; vm ) G(vm ) (vm v0 ){
(A.9)  h for m = 1> 2> = = = > q
](w; v0 ) G(v0 )
where { is the amount of the parallel shift. The level and swap rate Uv are given by
q
X G(vm )
O(w)
(A.10a) = m h(vm v0 ){
](w; v0 ) m=1 G(v0 )

G(v0 )  G(vq )h(vq v0 ){

(A.10b) Uv (w) = Xq =
m G(vm )h(vm v0 ){
m=1

Turning this around,

q
X
(A.11a) Uv m G(vm )h(vm v0 ){ + G(vq )h(vq v0 ){ = G(v0 )
m=1

determines the parallel shift { implicitly in terms of the swap rate Uv . With { determined by Uv , the level
is given by

O(Uv ) G(v0 )  G(vq )h(vq v0 ){

(A.11b) =
](w; v0 ) G(v0 )Uv
in terms of the swap rate. Thus this model yields

](w; ws ) Uv h(ws v0 ){

(A.12a) J(Uv ) =  >
O(w) G(vq ) (vq v0 ){
1 h
G(v0 )
where { is determined implicitly in terms of Uv by
q
X
(A.12b) Uv m G(vm )h(vm v0 ){ + G(vq )h(vq v0 ){ = G(v0 )=
m=1

This model’s limitations are that it allows only parallel shifts of the yield curve and it presumes perfect
correlation between long and short term rates.
A.4. Model 4: Non-parallel shifts. We can allow non-parallel shifts by approximating
](w; vm ) G(vm ) [k(vm )k(v0 )]{
(A.13)  h >
](w; v0 ) G(v0 )
where { is the amount of the shift, and k(v) is the eect of the shift on maturity v. As above, the shift { is
determined implicitly in terms of the swap rate Uv via
q
X
(A.14a) Uv m G(vm )h[k(vm )k(v0 )]{ + G(vq )h[k(vq )k(v0 )]{ = G(v0 )=
m=1
10
Then
O(Uv ) G(v0 )  G(vq )h[k(vq )k(v0 )]{
(A.14b) =
](w; v0 ) G(v0 )Uv

determines the level in terms of the swap rate. This model then yields

](w; ws ) Uv h[k(ws )k(v0 )]{

(A.15a) J(Uv ) =  >
O(w) G(vq ) [k(vq )k(v0 )]{
1 h
G(v0 )

where { is determined implicitly in terms of Uv by

q
X
(A.15b) Uv m G(vm )h[k(vm )k(v0 )]{ + G(vq )h[k(vq )k(v0 )]{ = G(v0 )=
m=1

To continue further requires selecting the function k(vm ) which determines the shape of the non-parallel shift.
This is often done by postulating a constant mean reversion,
1h i
(A.16) k(v)  k(v0 ) = 1  h(vv0 ) =

Alternatively, one can choose k(vm ) by calibrating the vanilla swaptions which have the same start date v0
and varying end dates to their market prices.

11