OVERNIGHT INDEXED SWAPS AND FLOORED COMPOUNDED

INSTRUMENT IN HJM ONE-FACTOR MODEL

MARC HENRARD

Abstract. Two types of financial instruments including (overnight) compounding are studied in
this note. The first one is overnight compounded instruments in the case where the settlement is
delayed with respect to the end of the compounding period (floating leg of the OIS). The second
is options on the composition. In both cases we study both continuous and discrete composition.
We provide explicit formulas within the HJM one-factor models with deterministic volatility
together with hedging strategies.

1. Introduction
Financial products based on composition of an overnight or short-term rate are fairly common.
In particular EONIA swaps (EUR) and overnight indexed swaps (OIS) based on Fed Fund Effective
rates (USD) are very liquid.
This note considers two types of instruments linked to composition. The first one is overnight
composition (floating leg of the OIS). This can appear obvious as the product is linear and can be
priced using the forward rates. This is true except that usually the payment of the amount due
is done one or two business days after the end of the composition. This extra time is necessary
to insure a smooth settlement of the transaction. From a valuation point of view, it means that
there is a so called convexity adjustment (the payment of interests does not take place at the end
of the period for which they are valid). We propose an explicit formula for the valuation of the
instrument in the case of continuous (Section 4) and discrete composition (Section 5). We also
estimate the difference between this option-like valuation and the standard valuation using forward
rates (Section 6).
The second subject we study about compounded products is the options on the composition.
The typical product we want to price is a bond paying the compounded average rate with a minimal
fixed rate. As the composition is some kind of average of the short term rates over the composition
period, this is an Asian option. We study this problem in the case of continuous (Section 7) and
discrete composition (Section 8) without payment lag or with payment lag (Section 9 and 10). For
these products we also obtain explicit formulas. The instrument described allow to invest at the
very short term rate, and so not taking too much interest rate risk, but with the guaranty of a
minimal return over the period, even if the rates reach very low levels. We also propose a version of
the formula for the case where the fixing of the rate is done in advance of the compounding period.
This is typically the case for Libor linked products in USD and EUR where the spot lag is two
days. The formulas of Section 7 and 8 can also be used to value option on the one day Brazilian
Inter-financial Deposits Index as negotiated on the BMF in Sao Paulo. Some characteristics of
this last instrument together with a valuation formula for the Hull-White volatility structure are
presented in Viera and Valls [8].
The most practically relevant sections are Section 9 for the options indexed on overnight rates
and Section 11 for Libor related products.

Date: First version: April 24, 2003; this version: February 3, 2004.
Key words and phrases. Overnight indexed swaps, option on OIS, Asian option, compounded average, explicit
formula, HJM model, one factor model, hedging.
JEL classification: G13, E43.
AMS mathematics subject classification:
1
2 M. HENRARD

The framework of this article is a one-factor Heath-Jarrow-Morton model with deterministic

volatility. The exact description of what we mean by that is done in the next section.

2. Model and hypothesis

We use a model for P (t, u), the price in t of the zero-coupon bond paying 1 in u. We will
describe this for all 0 ≤ t, u ≤ T , where T is some fixed constant.
When the discount curve P (t, .) is absolutely continuous, which is something that is always the
case in practice as the curve is constructed by some kind of interpolation, there exists f (t, u) such
that
 Z u 
(1) P (t, u) = exp − f (t, s)ds .
t

The idea of Heath-Jarrow-Morton [1] was to exploit this property by modeling f with a stochastic
differential equation
df (t, u) = µ(t, u)dt + σ(t, u)dWt
for some suitable (stochastic) µ and σ and deducing the behavior of P from there.
Here we use a similar model, but we restrict ourself to deterministic coefficients. We don’t need
all the technical refinement to create such a model (see for example the chapter on dynamical term
structure model in [4]). So instead of describing the conditions that lead to such a model, we
suppose that the conclusion of such a model are true. By this we mean we have a model, that we
call a HJM one-factor model, with the following property.
Let A = {(s, u) ∈ R2 : u ∈ [0, T ] and s ∈ [0, u]}. We work in a filtered probability space
(Ω, {Ft }, F, Preal ). The filtration Ft is the (augmented) filtration of a one-dimensional standard
Brownian motion (W real )0≤t≤T .

H: There exists σ : [0, T ]2 → R+ measurable and bounded1 with σ = 0 on [0, T ]2 \A such that
Rt
for some process (rs )0≤t≤T , Nt = exp( 0 rs ds) forms with some measure N a numeraire
pair2 (with Brownian motion Wt ),
Z u
df (t, u) = σ(t, u) σ(t, s)ds dt − σ(t, u)dWt
t
Z u
dP N (t, u) = P N (t, u) σ(t, s)ds dWt
t

and rt = f (t, t).

The notation P N (t, s) designates the numeraire rebased value of P , i.e. P N (t, s) = Nt−1 P (t, s).
To simplify the writing in the rest of the paper, we will use the notation
Z u
ν(t, u) = σ(t, s)ds.
t

Note that ν is increasing in u, measurable and bounded. Moreover for t > u, ν(t, u) = 0.
The following equations satisfied by the numeraire and the bonds will be useful in the rest of
the note:
dP (t, s) = P (t, s)rt dt + P (t, s)ν(t, s)dWt , dNt = Nt rt dt and dNt−1 = −Nt−1 rt dt.

1Bounded is too strong for the proof we use, some L1 and L2 conditions are enough, but as all the examples we
present are bounded, we use this condition for simplicity.
2See [4] for the definition of a numeraire pair. Note that here we require that the bonds of all maturities are
martingales for the numeraire pair (N, N).
 OIS AND FOCI 3

3. Preliminary results
We want to price some option in this model. For this we recall the generic pricing theorem [4,
Theorem 7.33-7.34].
Theorem 1. Let VT be some FT -measurable random Rt variable. If VT is attainable, then the time-t
value of the derivative is given by VtN = V0N + 0 φs dPsN where φt is a strategy and
Vt = Nt EN VT NT−1  Ft .
  

We now state two technical lemmas that generalize the lemmas presented in [2].
Lemma 1. Let 0 ≤ t ≤ u ≤ v. In a HJM one factor model, the price of the zero coupon bond can
be written has,
Z u
1 u 2
Z 
P (t, v) 2


P (u, v) = exp (ν(s, v) − ν(s, u)) dWs − ν (s, v) − ν (s, u) ds .
P (t, u) t 2 t
Proof. By definition of the forward rate and its equation,
 Z v 
P (u, v) = exp − f (u, τ )dτ
 Zuv  Z u Z u  
= exp − f (t, τ ) + ν(s, τ )D2 ν(s, τ )ds − D2 ν(s, τ )dWs dτ .
u t t

Then using again the definition of forward rates and the Fubini theorem on inversion of iterated
integrals, we have
 Z uZ v Z uZ v 
P (t, v)
P (u, v) = exp − ν(s, τ )D2 ν(s, τ )dτ ds + D2 ν(s, τ )dτ dWs
P (t, u) t u t u
 Z u Z u 
P (t, v) 1 2 2


= exp − ν (s, v) − ν (s, u) ds + ν(s, v) − ν(s, u)dWs .
P (t, u) 2 t t


Lemma 2. Let 0 ≤ u ≤ v. In the HJM one factor model, we have
 Z v Z v
1 v 2
 Z 
Nu Nv−1 = exp − rs ds = P (u, v) exp ν(s, v)dWs − ν (s, v)ds .
u u 2 u
Proof. By definition of r,
Z τ
rτ = f (τ, τ ) = f (t, τ ) + df (s, τ )ds
t
Z τ Z τ
= f (t, τ ) + ν(s, τ )D2 ν(s, τ )ds + D2 ν(s, τ )dWs .
t t

Then using Fubini, we have

Z v Z v Z vZ v Z vZ v
r(τ )dτ = f (t, τ )dτ + ν(s, τ )D2 ν(s, τ )dτ ds − D2 ν(s, τ )dτ dWs
u
Zuv u
Z vs Z v u s
1
= f (t, τ )dτ + ν 2 (s, v)ds + ν(s, v)dWs .
u 2 u u

4. Overnight indexed notes with continuous compounding

The first instrument we study is a note that pays, with a lag, the compounded short-term
interest, which is equivalent to the floating leg of an OIS.
4 M. HENRARD

Theorem 2. Let 0 ≤ t1 ≤ t2 ≤ t3 . In the a HJM one-factor model, the price of an instruments

that pays in t3 a principal of 1 and the short-term rate continuously compounded between t1 and
t2 (Nt−1
1
Nt2 ) is given at 0 by
P (0, t1 )
P (0, t3 ) β
P (0, t2 )
where Z t2 
β = exp (ν(s, t2 ) − ν(s, t1 ))(ν(s, t2 ) − ν(s, t3 ))ds .
0

Remark: For the Hull and White volatility model [3] with ν(s, t) = (1 − exp(−a(t − s)))σ/a, the
parameters β is given through
σ2
ln β = (exp(−at3 ) − exp(−at2 )) (cosh(at2 ) − cosh(at1 )) .
a3
Proof. Using the generic pricing Theorem 1 and the notation
Z t2
α2 = (ν(s, t1 ) + ν(s, t3 ) − ν(s, t2 ))2 ds.
t

we have  Z t2  
P (t2 , t3 )Nt−1 = EN Nt−1


V0 = EN exp rs ds 2 1
P (t2 , t3 ) .
t1
Using Lemma 1 with u = t2 and v = t3 and Lemma 2 with u = 0 and v = t1 , we have
  
P (0, t3 ) 1
V0 = P (0, t1 ) EN exp − α2 + αX β
P (0, t2 ) 2
where X is a standard normal distribution. Note that we use the fact that ν(s, t1 ) = 0 for s > t1
to write the integral between 0 and t1 as an integral between 0 and t2 .
Due to the fact that the expected value of the exponential of a normal distribution is given by
exp(µ + 21 σ 2 ) where µ is its average and σ its standard deviation, we have the result. 
Remark: If the payment takes place at the end of the interest period (t2 = t3 ), there is no convexity
adjustment and β = 1.
Remark: A payer OIS pays at settlement date (t3 ) a fixed interest against receiving the interest
continuously compounded between the start date (t1 ) and the end date (t2 ). The payment at
settlement is (Nt−11
Nt2 − 1) − R where R is the fixed interest amount. The value at 0 of the
instrument is  
P (0, t1 )
P (0, t3 ) β−1−R .
P (0, t2 )

5. Notes with discrete compounding

We now study the same instrument but with a discrete composition.
Theorem 3. Let t0 = 0 ≤ t1 < t2 < · · · < tn−1 ≤ tn . In the a HJM one-factor model, the price
of an instruments that pays in tn a principal ofQ 1 and the discrete compounding of interest rates
n−2
over the periods [ti , ti+1 ] (i = 1, . . . , n − 2) (i.e. i=1 P −1 (ti , ti+1 )) is given in 0 by
P (0, t1 )
P (0, tn ) βd
P (0, tn−1 )
where !
n−2
X Z ti
βd = exp − (ν(s, tn−1 ) − ν(s, ti ))(ν(s, tn ) − ν(s, tn−1 ))ds .
i=1 ti−1
 OIS AND FOCI 5

Remark: For the Hull and White volatility model [3], one has
n−2
σ2 X
ln βd = 3
(exp(−atn ) − exp(−atn−1 )) (exp(−ati ) − exp(−atn−1 )) (exp(2ati ) − exp(2ati−1 )) .
a i=1

Proof. Using the generic pricing theorem 1 we have

n−2
!
Y
V 0 = EN P −1 (ti , ti+1 )Nt−1
n
.
i=1
By Lemma 1,
n−2
Y P (0, t1 )
P −1 (ti , ti+1 ) =
i=1
P (0, tn−1 )
n−2 Z n−2
!
1 X ti 2 X Z ti
exp ν (s, ti+1 ) − ν 2 (s, ti )ds − ν(s, ti+1 ) − ν(s, ti )dWs
2 i=1 0 i=1 0

By splitting the integrals on the different sub-intervals [ti , ti+1 ] and rearranging the terms, we have
that
n−2
X Z ti n−2
X Z ti
ν(s, ti+1 ) − ν(s, ti )dWs = ν(s, tn−1 ) − ν(s, ti )dWs
i=1 0 i=1 ti−1

and a similar result for the other sum.

On the other hand, using the Lemma 2, we have
Z tn
1 tn 2
Z 
−1
Ntn = P (0, tn ) exp ν(s, tn )dWs − ν (s, tn )ds .
0 2 0
Let µ be defined on [0, T ] by

ν(s, tn ) − ν(s, tn−1 ) + ν(s, ti ) if s ∈ [ti−1 , ti ) i = 1, . . . , n − 1
µ(s) =
ν(s, tn ) if s ∈ [tn−1 , tn ]
Using those results and notations, we obtain
Z tn
1 tn 2
 Z 
P (0, t1 )
V0 = P (0, tn ) EN exp µ(s)dWs − µ (s)ds βd .
P (0, tn−1 ) 0 2 0

Remark: Like for the continuous composition, if the payment takes place at the end of the interest
period (tn−1 = tn ), there is no convexity adjustment and βd = 1.
Remark: Suppose that ν is continuous in (t, t) uniformly for t ∈ [0, tn ]. Let β be the number
defined in Theorem 2 on continuous compounding. Then
lim |βd − β| = 0
maxi=0,...,n−2 (ti+1 −ti )→0

This means that if all the intervals [ti , ti+1 ] are small, one can use the continuous version of the
instrument as an approximation of the discretely compounded version of the valuation. Numerical
estimates in the case of daily composition is given in the next section.

6. Comparison between discounted, continuously compounded and discretely

compounded HJM valuation of OIS
The difference between the option continuously compounded and forward approaches for an
overnight index swap is
P (0, t1 )
P (0, t3 ) (β − 1).
P (0, t2 )
We compute this difference in the case of the Hull-White volatility structure. Note that as
t1 ≤ t2 ≤ t3 and ν is increasing in the second variable, β < 1 and the difference is negative.
6 M. HENRARD

This is done with a flat curve (1.25%, ACT/360) which is more or less the shape of the USD
curve at the time of writing (March 2003).
Note that the yield curve play a role only through the discount factor between t2 and t3
(P (0, t3 )/P (0, t2 )) and the discount factor between 0 and t1 . So its role is minimal.
Usually, ti < 1 and a is small (typically between 0.01 and 0.10). We can then approximate the
exp and the cosh with their second order Taylor approximation. This gives
1  a 
ln β ≈ σ 2 (t2 − t3 ) − (t22 − t23 ) (t22 − t21 ).
2 2
As a(t22 − t23 ) is very small, the factor which has the largest influence on the adjustment is the
volatility factor σ.
Using the Taylor approximation of the exp and removing the factor in a we obtain
1 2
β−1≈ σ (t2 − t3 )(t22 − t21 ).
2
This means that the adjustment is in square of σ. Using standard market conventions, t1 is small
and we obtain that β − 1 ≈ 21 σ 2 (t3 − t2 )t2 2 .
The difference, using a mean reversion factor a = 0.01 and a volatility factor σ = 0.01 are given
in Table 1. We take t1 = 0, t2 equal 1w, 1m, 3m, 6m and 12m and t3 be 1, 2 or 7 days after t2 .
The difference is computed on a nominal of USD 1bn.

Compounding period
1w 1m 3m 6m 12m
lag 1d -0.15 -0.99 -9.06 -35.01 -137.08
2d -0.20 -1.97 -18.12 -70.01 -274.10
1w -0.35 -6.91 -63.40 -244.88 -958.65
Table 1. Difference between the option and forward valuations of the OIS.

We now look at the difference between the continuous and the discrete daily composition. For
this comparison, we use 31 December 2002 as value date. With that date, the spot date (t1 ) is 3
January 2003 and the 6 months maturity is 3 July 2003. So the payment dates are respectively 7,
8 and 10 July. This increases the difference between the end of compounding period and payment
date. Even with this the difference in valuation is very small. The results are given in Table 2

Compounding period
1w 1m 3m 6m 12m
lag 1d 0.04 0.07 0.18 1.46 0.75
2d 0.05 0.14 0.74 1.82 1.49
1w 0.09 0.48 1.29 2.55 5.22
Table 2. Difference between the daily discrete composition valuation and the
continuous one of the OIS.

As a conclusion on this section we can say that, for practical purposes, the difference between the
forward valuation of OIS and its valuation using option approach, in its continuous compounding
version or daily discretely compounding one, can be neglected. For example with the standard
settlement of two days, the maximal difference is around 250 for a one year OIS with a notional
of 1 bn. This is approximatively the error coming from a discrepancy of 1/400 bp on the one year
rate for the fixed leg of the swap. Even if the volatility is doubled and the adjustment quadrupled,
it stays extremely small.
 OIS AND FOCI 7

7. Floored and capped continuously compounded instruments

In this section, we analyze options on compounded instruments in the absence of spot-lag (t1 = 0,
t2 = t3 ). We give the pricing formula at any time between the start and the end of the composition.
This allows us to describe the hedging strategy in the second theorem of the section.
Theorem 4. Let 0 ≤ t < T and K > 0. In the a HJM one-factor model with deterministic
volatility, the price of an instrument paying in T the maximum of a fixed amount K and the sum
of a principal of 1 and the short-term rate continuously compounded between 0 and T (max(K, NT ))
(floored instrument) is given in t by
Ft = Nt N (κ) + KP (t, T )N (σ − κ)
where    
1 Nt 1
κ= ln + σ2
σ KP (t, T ) 2
and Z T
σ2 = ν 2 (s, T )ds.
t
The price of the instrument paying the minimum of K and the compounding (capped instrument)
(min(K, NT )) is given in t by
Ct = Nt N (−κ) + KP (t, T )N (κ − σ).
Remark: For the Hull and White volatility structure,
σ2
 
2 3 2 1
σ = 2 T −t− + exp(−a(T − t)) − exp(−2a(T − t)) .
a 2a a 2a
Proof. By Lemma 2, we have
Z T ! !
Z T Z T
1
NT = Nt exp rs ds = Nt P −1 (t, T ) exp − ν(s, T )dWs + 2
ν (s, T )ds .
t t 2 t

As the stochastic integral of a non-stochastic function is normal ([5, Section 3.6, p 65] or [7,
Theorem 3.1, p. 60]) we can write
 
−1 1 2
NT = Nt P (t, T ) exp σ − σX
2
with X a standard normal distribution.
The asset NT is larger than K if and only if X < κ, so for the floored instrument we have
Ft = Nt EN max(NT , K)NT −1  Ft
  
    
1 2 
= Nt PN [ X < κ| Ft ] + KP (t, T ) EN exp − σ + σX 11(X ≥ κ) Ft .
2
In the case of the capped instrument, we have
    
1 2 
Ct = Nt PN [ X > κ| Ft ] + KP (t, T ) EN exp − σ + σX 11(X ≤ κ) Ft .
2
As X is independent of Ft and κ is Ft -measurable, using a property of the conditional expec-
tation [5, Proposition A.2.5], we have that VtF = Nt φ1 (κ) + KP (t, T )φ2 (κ) where
   
1
φ1 (y) = P (X < y) and φ2 (y) = EN exp − σ 2 + σX 11(X ≥ y) .
2
So we obtain
Ft = Nt N (κ) + KP (t, T )N (σ − κ).
Similarly
Ct = Nt N (−κ) + KP (t, T )N (κ − σ).
8 M. HENRARD


We turn now to the hedging strategy.
Theorem 5. Under the hypothesis of Theorem 4, a hedging strategy for the floored instrument is
to hold a nominal
∆t = KN (σ − κ)
of the bond P (t, T ) and an amount of Nt N (κ) in the cash account (N (κ) units of the numeraire).
Proof. Using the results of the previous Theorem, we know that the value can be written as
FtN = G(t, P (t, T ), Nt )Nt−1 .
Using the multidimensional Itô formula and writing Xs1 = P (s, T ) and Xs2 = Ns , we have
Z t Z t
−1
FtN
= F0 + N
G(s, P (s, T ), Ns )dNs + Ns−1 D1 G(s, P (s, T ), Ns )ds
0 0
Z t Z t
+ Ns−1 D2 G(s, P (s, T ), Ns )dP (s, T ) + Ns−1 D3 G(s, P (s, T ), Ns )dNs
0 0
Z t 2
1 X
Ns−1 2
G(s, P (s, T ), Ns )d X i , X j s


+ Di+1,j+1
2 0 i,j=1
Z t Z t
= F0N + D2 F (s, P (s, T ), Ns )dP N (s, T ) + Hs ds.
0 0
As FtN is a martingale under N we have Hs = 0. So we obtain
Z t
N N
Ft = F0 + D2 G(s, P (s, T ), Ns )dP N (s, T ).
0
By Theorem 1, we have that the hedging quantity in the bond of maturity T is
∆t = D2 G(t, P (t, T ), Nt ).
We now compute the value of this derivative. For this remember that
G(t, P (s, T ), Nt ) = Nt N (κ) + KP (t, T )N (σ − κ)
where κ is implicitly defined by
 
1 2
g(P, N, κ) = Nt P −1 (t, T ) exp σ − σκ − K = 0.
2
We use the implicit function theorem [6]. As the derivative of g with respect to κ and P exist and
the first one is non zero, the derivative of κ with respect to P exists. So we have
D2 G(t, P (t, T ), Nt ) = Nt N 0 (κ)D1 κ − KP (t, T )N 0 (σ − κ)D1 κ + KN (σ − κ)
   
1 1 1
= D1 κ √ exp( − κ2 Nt − KP exp σκ − σ 2 + KN (σ − κ)
2π 2 2
= KN (σ − κ).


8. Floored and capped discretely compounded instruments

This section is devoted to the pricing and hedging of instruments with discrete composition.
Theorem 6. Let 0 = t0 ≤ t < t1 < t2 < · · · < tn and K > 0. In the a HJM one factor model, the
price of an instrument paying in tn the maximum of a fixed amount K and the sum of a principal
of 1 and the discrete compounding of interest rates over the periods [ti , ti+1 ] (i = 0, . . . , n − 1) (i.e.
Qn−1 −1
i=0 P (ti , ti+1 )) is given in t by
P (t, t1 )
Ft = N (κ + σ) + KP (t, tn )N (−κ)
P (0, t1 )
 OIS AND FOCI 9

where
n−1
X Z ti 2
σd2 = (ν(s, tn ) − ν(s, ti )) ds
i=1 t∨ti−1

and    
1 P (t, t1 ) 1
κd =
ln − σd2 .
σd KP (0, t1 )P (t, tn ) 2
The price of an instrument paying in tn the minimum of K and the discrete composition (capped
instrument) is given in 0 by
P (t, t1 )
C0 = N (−κ − σ) + KP (t, tn )N (κ).
P (0, t1 )
Remark: For the Hull and White volatility structure,
n−1
σ2 1 1 X
σd2 = exp(−2atn )(exp(−2atn−1 ) − exp(−2at)) + (1 − exp(−2a(ti − t ∨ ti−1 )))
a2 2a 2a i=1
n−1
!
X
−2 exp(−atn ) exp(−ati )(ti − t ∨ ti−1 ) .
i=1

Proof. The price of the instrument is

(n−1 ) !
Y
Ft = Nt EN max P −1 (ti , ti+1 ), K Nt−1
n
.
i=0

Using Lemma1, we have

n−1
Y P (t, t1 )
P −1 (ti , ti+1 ) =
i=0
P (0, t1 )P (t, tn )
n−1 Z n−1
!
1 X ti 2 X Z ti
exp ν (s, ti+1 ) − ν 2 (s, ti )ds − ν(s, ti+1 ) − ν(s, ti )dWs
2 i=1 t i=1 t

By splitting the integrals on the different sub-intervals [ti , ti+1 ] and rearranging the terms, we have
that
n−1
X Z ti n−1
X Z ti
ν(s, ti+1 ) − ν(s, ti )dWs = ν(s, tn ) − ν(s, ti )dWs
i=1 t i=1 t∨ti−1

(where t ∨ s is the maximum between t and s) and a similar result for the other sum.
By Lemma 2,
Z tn
1 tn 2
Z 
Nt Nt−1 = P (t, tn ) exp ν(s, t n )dW s − ν (s, t n )ds .
n
t 2 t
Rs
We denote this last exponential by Ltn . Let Ws# = Ws − 0 ν(τ, tn )dτ . By the Girsanov’s theorem
([5, Section 4.2.2, p. 72]), Wt# is a standard Brownian motion with respect to the probability P#
of density Ltn with respect to N.
The value of the instrument can now be written as
    
P (t, t1 ) 1
Ft = E# P (t, tn ) max exp − σd2 − σd X # , K
P (0, t1 )P (t, tn ) 2
where X # is a random variable with a standard normal distribution with respect to P# . The first
term of the maximum operator is the actual maximum when X # < κd . So we obtain
   
P (t, t1 ) # 1
Ft = E exp −σd X # − σd2 11(X # < κd ) + KP (t, tn )P# (X # ≥ κd )
P (0, t1 ) 2
10 M. HENRARD

which by standard manipulation on the expectation and on the normal distribution lead to the
result.
The proof for the capped instrument is similar. 
We turn now to the hedging strategy.
Theorem 7. Under the hypothesis of Theorem 6, a hedging strategy for the floored instrument is
to hold a nominal of ∆1t of the bond P (t, t1 ) and a nominal ∆2t of the bond P (t, tn ) with
1
∆1t = N (κ + σ)
P (0, t1 )
and
∆2t = KN (−κ)
(and no cash).
Proof. Using the results of the previous Theorem, we know that the option value can be written
as
FtN = G(P (t, t1 ), P (t, tn ))Nt−1 .
Using the multidimensional Itô formula and writing Xs1 = P (s, t1 ) and Xs2 = P (s, tn ), we have
Z t Z t
FtN = F0N + G(Xs1 , Xs2 )dNs−1 + Ns−1 D1 G(Xs1 , Xs2 )dP (s, t1 )
0 0
t 2
1 t −1 X 2
Z Z
Ns−1 D2 G(Xs1 , Xs2 )dP (s, tn ) Di,j G(Xs1 , Xs2 )d X i , X j s


+ + Ns
0 2 0 i,j=1
Z t Z t Z t
= F0N + D1 G(Xs1 , Xs2 )dP N (s, t1 ) + D2 G(Xs1 , Xs2 )dP N (s, tn ) + Hs ds.
0 0 0

As FtNis a martingale under N we have Hs = 0. By Theorem 1, we have that the hedging quantity
in the bond of maturity t1 and tn are
∆1t = D1 G(Xt1 , Xt2 ) and ∆2t = D2 G(Xt1 , Xt2 ).
We now compute the value of this derivative. For this remember that
P (t, t1 )
G(P (t, t1 ), P (t, tn )) = N (κ + σ) + KP (t, tn )N (−κ)
P (0, t1 )
where κ is implicitly defined by
 
1 2 −1 1 2
g(X , X , κ) = P (0, t1 )X exp − σ − σκ − KX 2 = 0.
1
2
We use the implicit function theorem [6]. As the derivative of g with respect to κ, X 1 and X 2
exist and the first one is non zero, κ can be written locally as a function of X 1 and X 2 and the
derivatives of the implicit function exist. So we have
X1
D2 G(Xt1 , Xt2 ) = KN (−κ) + N 0 (κ + σ)D2 κ − KX 2 N 0 (κ)D2 κ
P (0, t1 )
X1
   
1 1 2 1 2 2
= KN (−κ) + D2 κ √ exp(− κ ) exp − (2κσ + σ ) − KX
2π 2 P (0, t1 ) 2
= KN (−κ),
and similarly
1
D1 (Xt1 , Xt2 ) = N (κ + σ).
P (0, t1 )

 OIS AND FOCI 11

9. Floored and capped continuously compounded instruments with payment lag

We come back to the pricing of options on compounded instrument, but this time in the presence
of payment lag.
Theorem 8. Let 0 ≤ t ≤ t1 ≤ t2 ≤ t3 , K > 0,
Z t2 Z t2 Z t2
2 2
σ1 = (ν(s, t2 ) − ν(s, t1 )) ds, σ12 = ν(s, t3 ) (ν(s, t2 ) − ν(s, t1 )) ds, σ22 = ν 2 (s, t3 )ds
t t t
and the matrix Σ defined by
σ12 σ12
 
Σ= .
σ12 σ22
In the a HJM one-factor model with deterministic volatility, if the matrix Σ is invertible, the
price of an instrument paying in t3 the maximum of a fixed amount K and the principal Q gross-up
by the short-term rate continuously compounded between t1 and t2 (QNt−1 1
Nt2 ) (floored instrument
with spot lag) is given in t by
σ 2 − σ12
   
P (t, t1 ) σ12
Ft = QP (t, t3 ) βN κ + 1 + KP (t, t3 )N −κ
P (t, t2 ) σ1 σ1
where    
1 P (t, t1 ) 1
κ= ln Q − σ12 + α ,
σ1 KP (t, t2 ) 2
Z t2
α= ν(s, t2 )(ν(s, t2 ) − ν(s, t1 ))ds
t
and Z t2 
β = exp (ν(s, t2 ) − ν(s, t1 ))(ν(s, t2 ) − ν(s, t3 ))ds .
t
The price of an instrument paying in t3 the minimum of K and the composition between t1 and
t2 (capped instrument with spot lag) is given in 0 by
σ12 − σ12
   
P (t, t1 ) σ12
Ct = QP (t, t3 ) βN −κ − + KP (t, t3 )N κ −
P (t, t2 ) σ1 σ1
Proof. Using the generic pricing theorem 1 we have
Ft = Nt EN max QNt−1 Nt2 , K P (t2 , t3 )Nt−1
   
 Ft .
1 2

This expected value can be computed explicitly using standard decomposition and computation
of normal distribution. Here those computation are a little bit more involved and required some
extra notations. Let
Z t2 Z t2
X1 = ν(s, t2 ) − ν(s, t1 )dWs , X2 = ν(s, t3 )dWs .
t t
The random variables X1 and X2 are jointly normally distributed ([7, Theorem 3.1, p. 60]) with
covariance matrix Σ.
With those notations, using Lemma 2 twice and the fact that ν(s, t1 ) = 0 for s > t1 , we have
 
−1 P (t, t1 ) 1 2
Nt1 Nt2 = exp −X1 − σ1 + α
P (t, t2 ) 2
and so QNt−1
1
Nt2 > K when X1 < σ1 κ. Moreover, using Lemmas 1 and 2,
 
−1 1 2
Nt Nt2 P (t2 , t3 ) = P (t, t3 ) exp X2 − σ2
2
and  
P (t, t3 ) 1 1
Nt Nt−1 P (t2 , t3 ) = P (t, t1 ) exp X2 − X1 − σ22 − σ12 + α .
1
P (t, t2 ) 2 2
12 M. HENRARD

So the expected value is obtain by

Z    
P (t, t3 ) 1 1 1
Ft = A QP (t, t1 ) exp x2 − x1 − σ22 − σ12 + α exp − xT Σ−1 x dx
x1 <σ1 κ P (t, t2 ) 2 2 2
Z    
1 1
+A KP (t, t3 ) exp − σ22 + x2 exp − xT Σ−1 x dx
x1 ≥σ1 κ 2 2
1 √1 .
where A = 2π |Σ|
To obtain an explicit solution, we will need to compute the following integral
Z  
1 1
I=√ exp x2 − xT Σ−1 x dx2 .
2π R 2
For that we use the notation  2 
τ1 τ12
T = Σ−1 = .
τ12 τ22
and we obtain   
1 1 |T | 2 τ12 1
I= exp − x + 2x 1 − .
τ2 2 τ22 1 τ22 τ22
So we are now left with one dimensional integrals in the computation of the price
Z   2 !
P (t, t3 ) 1 1 1 τ12 1
Ft = QP (t, t1 ) exp(α) √ exp − x1 + σ1 2 +1 − σ12 dx1
P (t, t2 ) 2π x1 <σ1 κ 2 σ1 τ2 σ1
Z  2 !
1 1 1 σ12 1
+KP (t, t3 ) √ exp − x1 − dx1 .
2π x1 ≥σ1 κ 2 σ 1 σ 1 σ 1

A straightforward but tedious computation gives then the result. In particular we use the fact that
exp(α − σ12 ) = β.
The proof for the capped instrument is similar. 
Remark: The coefficient β is the same as the one defined in Theorem 2.
If t1 = 0 and t2 = t3 the matrix Σ is not invertible and the proof does not hold, but the formula
we obtain is the one of Theorem 4 and is still valid. For ν continuous, the value of the option is
continuous in t1 , t2 and t3 including in t1 = 0 and t2 = t3 .
Remark: If the composition period has already started (0 ≤ t¯1 < t), one can use the formula but
with t1 = t and a modified notional. The notional is modified by the interests already compounded:
Q = Q̄Nt−1 −1
¯1 Nt . The formula is similar with only Q replaced by Q̄Nt¯1 Nt . With this adjustment
one can price “aged” instruments for which the composition period has already started.
We turn now to the hedging strategy.
Theorem 9. Under the hypothesis of Theorem 8, a hedging strategy for the floored instrument is
to hold a nominal of ∆it of the bonds P (t, ti ) (i = 1, 2, 3) with
σ12 − σ12
 
1 P (t, t3 )
∆t = Q βN κ + ,
P (t, t2 ) σ1
σ 2 − σ12
 
P (t, t1 )
∆2t = −QP (t, t3 ) 2 βN κ + 1
P (t, t2 ) σ1
and  
σ12
∆3t = KN −κ .
σ1
(and some cash...).
Proof. Like in the proof of Theorem 7, we have that ∆it = Di G(Xs1 , Xs2 ).
We now compute the value of the derivatives. For this remember that
σ 2 − σ12
   
P (t, t1 ) σ12
G(P (t, t1 ), P (t, tn )) = QP (t, t3 ) βN κ + 1 + KP (t, t3 )N −κ
P (t, t2 ) σ1 σ1
 OIS AND FOCI 13

where κ is implicitly defined by

1
g(X 1 , X 2 , κ) = QP (t, t1 )β exp(−σ1 κ − σ12 + α) − KX 2 = 0.
2
Using again techniques similar to the one of Theorem 7 we obtain the result. 

10. Floored and capped discretely compounded instruments with payment lag
This section is of less practical importance. For overnight indexed instruments, the continuous
compounding is close enough to the discrete compounding for practical purposes. Instruments with
a longer reset tenor are usually linked to a Libor-like rate that fixes in advance and pays without
lag. Those last instruments are the subject of the next section.
Theorem 10. Let 0 ≤ t1 ≤ t2 ≤ · · · ... ≤ tn−1 ≤ tn , K > 0,
Pn−3 R t 2 Rt
σ12 = i=0 tii+1 (ν(s, tn−1 ) − ν(s, ti+1 )) ds, σ22 = 0 n−1 ν 2 (s, tn )ds,
Pn−3 R t
σ12 = i=0 tii+1 ν(s, tn ) (ν(s, tn−1 ) − ν(s, ti+1 )) ds
and the matrix Σ be defined by
σ12 σ12
 
Σ= .
σ12 σ22
In the a HJM one-factor model with deterministic volatility, if the matrix Σ is invertible, the
price of an instrument paying in tn the maximum of a fixed amount K and the sum of a principal
of 1 and the discrete compounding of interest rates over the periods [ti , ti+1 ] for i = 1, . . . , n − 2
Qn−2
(i.e i=1 P −1 (ti , ti+1 )) is given in 0 by
σ12 − σ12
   
P (0, t1 ) σ12
F0 = P (0, tn ) βN κ + + KP (0, tn )N −κ
P (0, tn−1 ) σ1 σ1
where    
1 P (0, t1 ) 1 2
κ= ln − σ1 + α ,
σ1 KP (0, tn−1 ) 2
n−3
X Z ti+1
α= ν(s, tn−1 )(ν(s, tn−1 ) − ν(s, ti+1 ))ds
i=0 ti

and !
n−3
X Z ti+1
β = exp (ν(s, tn−1 ) − ν(s, ti+1 ))(ν(s, tn−1 ) − ν(s, tn ))ds .
i=0 ti

The price of an instrument paying in tn the minimum of a fixed amount K and the sum of
a principal of 1 and the discrete compounding of interest rates over the periods [ti , ti+1 ] (capped
discrete instrument with spot lag) is given in 0 by
σ 2 − σ12
   
P (0, t1 ) σ12
C0 = P (0, tn ) βN −κ − 1 + KP (0, tn )N κ −
P (0, tn−1 ) σ1 σ1
Proof. Using the generic pricing theorem 1 we have
(n−2 ) !
Y
V0 = EN max P −1 (ti , ti+1 ), K P (tn−1 , tn )Nt−1
n−1
.
i=1

This expected value can be computed explicitly using standard decomposition and computation
of normal distribution. Here those computation are a little bit more involved and required some
extra notations. Let
n−3
X Z ti+1 Z tn−1
X1 = ν(s, tn−1 ) − ν(s, ti+1 )dWs , X2 = ν(s, tn )dWs .
i=0 ti 0
14 M. HENRARD

The random variables X1 and X2 are jointly normally distributed ([7, Theorem 3.1, p. 60]) with
covariance matrix Σ.
With those notations and using Lemma 1, we have
n−2  
Y P (0, t1 ) 1
P −1 (ti , ti+1 ) = exp −X1 − σ12 + α
i=1
P (0, tn−1 ) 2
Qn−2 −1
and so i=1 P (ti , ti+1 ) > K when X1 < σ1 κ. Moreover, using Lemmas 1 and 2,
 
−1 1 2
Ntn−1 P (tn−1 , tn ) = P (0, tn ) exp X2 − σ2
2
and
n−2  
Y P (0, t1 ) 1 2 1 2
P −1 (ti , ti+1 )Nt−1 P (t n−1 , t n ) = P (0, t n ) exp X 2 − X 1 − σ − σ + α .
i=1
n−1
P (0, tn−1 ) 2 2 2 1
So the expected value is obtain exactly like in the previous theorem, except the the constant
have slightly different meaning. 

11. Floored and capped discretely compounded instruments with fixing lag
Theorem 11. Let 0 ≤ t1 < t2 < · · · < tn , 0 = s0 ≤ t < s1 < s2 < · · · < sn−1 with si ≤ ti and
K > 0. In the a HJM one factor model, the price of an instrument paying in tn the maximum of
a fixed amount K and of a principal of Q gross-up by the discrete compounding of interest rates
Qn−1
over the periods [ti , ti+1 ] fixed in si (i.e. i=1 P (si , ti )/P (si , ti+1 )) is given in t by
Ft = QP (t, t1 )N (κ + σ) + KP (t, tn )N (−κ)
where
n−1
X Z si
2 2
σ = (ν(s, tn ) − ν(s, ti )) ds
i=1 t∨si−1
and    
1 P (t, t1 ) 1
ln κ= − σ2 .
σ KP (t, tn ) 2
The price of an instrument paying in tn the minimum of a fixed amount K and of a principal
of Q gross-up by the discrete compounding of interest rates over the periods [ti , ti+1 ] fixed in si is
given in t by
Ct = QP (t, t1 )N (−κ − σ) + KP (t, tn )N (κ)
Proof. The price of the instrument is
( n−1
) !
Y P (si , ti )
Ft = Nt EN max Q ,K Nt−1 .
i=1
P (si , ti+1 ) n

Using Lemma1, we have

n−1 n−1 Z n−1
!
Y P (si , ti ) P (t, t1 ) 1 X si 2 2
X Z si
= exp ν (s, ti+1 ) − ν (s, ti )ds − ν(s, ti+1 ) − ν(s, ti )dWs
i=1
P (si , ti+1 ) P (t, tn ) 2 i=1 t i=1 t
By splitting the integrals on the different sub-intervals [si , si+1 ] and rearranging the terms, we
have that
n−1
X Z si n−1
X Z si
ν(s, ti+1 ) − ν(s, ti )dWs = ν(s, tn ) − ν(s, ti )dWs
i=1 t i=1 t∨si−1
and a similar result for the other sum.
By Lemma 2,
Z tn Z tn 
1
Nt Nt−1 = P (t, tn ) exp ν(s, tn )dWs − ν 2 (s, tn )ds .
n
t 2 t
 OIS AND FOCI 15

Rs
We denote this last exponential by Ltn . Let Ws# = Ws − 0 ν(τ, tn )dτ . By the Girsanov’s theorem
([5, Section 4.2.2, p. 72]), Wt# is a standard Brownian motion with respect to the probability P#
of density Ltn with respect to N.
The value of the instrument can now be written as
    
# P (t, t1 ) 1 2 #
Ft = E P (t, tn ) max Q exp − σ − σX ,K
P (t, tn ) 2

where X # is a random variable with a standard normal distribution with respect to P# . The first
term of the maximum operator is the actual maximum when X # < κ. So we obtain
   
1
Ft = QP (t, t1 ) E# exp −σX # − σ 2 11(X # < κ) + KP (t, tn )P# X # ≥ κ


2
which by standard manipulation on the expectation and on the normal distribution lead to the
result.
The proof for the capped instrument is similar. 
Remark: The case where t is after the fist fixing (t ≥ s1 ) can be treated in a way similar to
Section 9. The formula can be applied with only the remaining fixing but the notional modified by
Qk
the interest already fixed. If sk ≤ t < sk+1 , the notional is replaced by Q i=1 P (si , ti )/P (si , ti+1 ).

Theorem 12. Under the hypothesis of Theorem 11, a hedging strategy for the floored instrument
is to hold a nominal of ∆1t of the bond P (t, t1 ) and a nominal ∆2t of the bond P (t, tn ) with

∆1t = QN (κ + σ) and ∆2t = KN (−κ)

(and no cash).

Proof. Like in the proof of Theorem 7, we have that ∆it = Di G(Xs1 , Xs2 ). And a similar argument
lead to ∆1t = QN (κ + σ) and ∆2t = KN (−κ). 

12. Conclusion
For overnight indexed swaps, the convexity adjustment due to the delay between the end of the
composition period and the payment is, for all practical purposes, negligible. Also the difference
between the continuous composition and the discrete composition can, for daily composition, be
neglected.
We propose analytical formulas for floor and cap on the compounded average rate. Those
formulas are proposed for continuous and discrete compounding and with and without spot-lag.
Like for the overnight indexed swaps, the price with discrete composition tends to the price with
continuous composition when the composition interval lengths tend to 0. Also the value with
spot-lag tends to the value without when the lag tends to 0.
Nevertheless the exact formula allows to see the exact dependency of the price on the rate of
different maturity and the volatility structure. For discrete composition, the price includes the
volatility to the start date of the last composition. So when the last interval is long, the difference
between continuous and discrete composition can be substantial.
We finally propose results for floored and capped instruments on composition with fixing lag.
This apply particularly for Libor related products.

Disclaimer: The views expressed here are those of the author and not necessarily those of the
Bank for International Settlements.
16 M. HENRARD

References
[1] D. Heath, R. Jarrow, and A. Morton. Bond pricing and the term structure of interest rates: a new methodology
for contingent claims valuation. Econometrica, 60(1):77–105, January 1992. 2
[2] M. Henrard. Explicit bond option and swaption formula in Heath-Jarrow-Morton one factor model. International
Journal of Theoretical and Applied Finance, 6(1):57–72, February 2003. 3
[3] J. Hull and A. White. Princing interest rate derivatives securities. The Review of Financial Studies, 3:573–592,
1990. 4, 5
[4] P. J. Hunt and J. E. Kennedy. Financial Derivatives in Theory and Practice. Wiley series in probability and
statistics. Wiley, 2000. 2, 3
[5] D. Lamberton and B. Lapeyre. Introduction au calcul stochastique appliqué à la finance. Ellipses, 1997. 7, 9, 15
[6] J. Mawhin. Analyse: Fondements, Techniques, Evolution. De Boek, 1997. 8, 10
[7] R. Rebonato. Modern pricing of interest-rate derivatives: the LIBOR market model and Beyond. Princeton
University Press, Princeton and Oxford, 2002. 7, 11, 14
[8] C. A. Viera and P. L. Valls Pereira. Closed form formula for the price of options on the one day Brazilian
Interfinancial Deposit Index - IDI. Working paper, IBMEC Business School, 2001. 1

Contents
1. Introduction 1
2. Model and hypothesis 2
3. Preliminary results 3
4. Overnight indexed notes with continuous compounding 3
5. Notes with discrete compounding 4
6. Comparison between discounted, continuously compounded and discretely compounded
HJM valuation of OIS 5
7. Floored and capped continuously compounded instruments 7
8. Floored and capped discretely compounded instruments 8
9. Floored and capped continuously compounded instruments with payment lag 11
10. Floored and capped discretely compounded instruments with payment lag 13
11. Floored and capped discretely compounded instruments with fixing lag 14
12. Conclusion 15
References 16

Derivatives Group, Banking Department, Bank for International Settlements, CH-4002 Basel (Switzer-
land)
E-mail address: Marc.Henrard@bis.org