Lecture Notes on Financial Mathematics

c Harald Lang, KTH Mathematics

°
version 15/5-07

Preface
The aim of these notes is to present (some of) the fundamental features of
financial mathematics in a rigorous way but avoiding stochastic calculus (stochastic
differential equations.) This means that time is discrete, and the continuous case
is considered as the limiting case when the time intervals → 0.
These notes are not intended to be a stand alone text on financial mathemat-
ics; rather, they are intended to be a mathematical supplement to some text on
financial economics. A good such text book is John C. Hull’s: Options, Futures,
and Other Derivatives (Prentice Hall.)

Harald Lang
Lecture Note 1

Note 1: Introduction to Present-, Forward-

and Futures Prices

Assume that we want to buy a quantity of coffee beans with delivery in nine
months. However, we are concerned about what the (spot) price of coffee beans
might be then, so we draw up a contract where we agree on the price today. There
are now at least three ways in which we can arrange the payment: 1) we pay now,
in advance. We call this price the present price of coffee beans with delivery in
nine months time, and denote it by P . Note that this is completely different from
the spot price of coffee beans, i.e., the price of coffee for immediate delivery. 2)
we pay when the coffee is delivered, i.e., in nine months time. This price is the
forward price, which we denote by G. 3) we might enter a futures contract with
delivery in nine months time. A futures contract works as follows:
Let us denote the days from today to delivery by the numbers 0, 1, . . . , T , so
that day 0 is today, and day T is the day of delivery. Each day n a futures price
Fn of coffee beans with delivery day T is noted, and FT equals the prevailing spot
price of coffee beans day T . The futures price Fj is not known until day j; it will
depend on how the coffee bean crop is doing, how the weather has been up to that
day and the weather prospects up till day T , the expected demand for coffee, and
so on. One can at any day enter a futures contract, and there is no charge for doing
so. The long holder of the contract will each day j receive the amount Fj − Fj−1
(which may be negative, in which he has to pay the corresponding amount,) so if
I enter a futures contract at day 0, I will day one receive F1 − F0 , day two F2 − F1
and so on, and day T , the day of delivery, FT − FT −1 . The total amount I receive
is thus FT − F0 . There is no actual delivery of coffee beans, but if I at day T buy
the beans at the spot price FT , I pay FT , get my coffee beans and cash the amount
FT − F0 from the futures contract. In total, I receive my beans, and pay F0 , and
since F0 is known already day zero, the futures contract works somewhat like a
forward contract. The difference is that the value FT − F0 is paid out successively
during the time up to delivery rather than at the time of delivery.
The simplest of these three contracts is the one when we pay in advance, at
least if the good that is delivered is non-pecuniary, since in that case the interest
does not play a part. For futures contrats, the interest rate clearly plays a part,
since the return of the contract is spread out over time.
We will derive some book-keeping relations between the present prices, for-
ward prices and futures prices, but first we need some interest rate securities.

Zero coupon bonds

A zero coupon bond with maturity T is a contract where the long holder pays
ZT day 0 and receives the amount 1 day T . One can take both long and short
positions on zero coupon bonds. There is nothing random about a zero coupon
bond, both the price ZT paid today and the amount 1 received day T is known
by day 0.

1
Lecture Note 1

Money Market Account

A money market account is like a series of zero coupon bonds, maturing after
only one day (or whatever periods we have in our futures contracts—it might be
for example a week.) If I deposit the amount 1 day 0, the balance of my account
day 1 is er1 , where r1 is the short interest rate from day 0 to day 1. The rate r1 is
known already day 0. The next day, day 2, the balance has grown to er1 +r2 , where
r2 is the short interest rate from day 1 to day 2; it is random as seen from day 0,
and its outcome is determined day 1. The next day, day 3, the balance has grown
to er1 +r2 +r3 , where r3 is the short interest rate from day 2 to day 3; it is random
as seen from day 0 and day 1, and its ouctome occurs day 2, and so on. Day T
the balance is thus er1 +···+rT which is a random variable. In order to simplify the
notation, we introduce the symbol R(t, T ) = rt+1 + · · · + rT .

Relations between Present-, Forward- and Futures Prices

Let P (T ) [X] be the present price of a contract that delivers the random value
X (which may take negative values) at time T . Likewise, let G(T ) [X] denote the
(T )
forward price of the value X delivered at time T and Ft [X] the futures price as
of time t of the value X delivered at time T . We then have the following theorem:

Theorem
The following relations hold:
a) P , G and F0 are linear functions, i.e., if X and Y are random payments
made at time T , then for any constants a and b P (T ) [aX + bY ] = aP (T ) [X] +
bP (T ) [Y ], and similarly for G and F0 .
(T )
b) G(T ) [1] = 1, F0 [1] = 1 and P (T ) [1] = ZT
c) P (T ) [X] = ZT G(T ) [X]
(T )
d) P (T ) [XeR(0,T ) ] = F0 [X]
(T )
e) P (T ) [X] = F0 [Xe−R(0,T ) ]

Proof
The proof relies on an assumption of the model: the law of one price. It
means that there can not be two contracts that both yield the same payoff X at
time T , but have different prices today. Indeed, if there were two such contracts,
we would buy the cheaper and sell the more expensive, and make a profit today,
and have no further cash flows in the future. But so would everyone else, and this
is inconsistent with market equilibrium. In Lecture Note 4 we will extend this
model assumption somewhat.
To prove c), note that if we take a long position on a forward contract on X
and at the same time a long position of G(T ) [X] zero coupon bonds, then we have
a portfolio which costs ZT G(T ) [X] today, and yields the income X at time T . By
the law of one price, it hence must be that c) holds.
To prove d), consider the following strategy: Deposit F0 on the money market
account, and take er1 long positions on the futures contract on X for delivery at
time T .

2
Lecture Note 1

The next day the total balance is then F0 er1 + er1 (F1 − F0 ) = F1 er1 . Deposit
this on the money market account, and increase the futures position to er1 +r2
contracts.
The next day, day 2, the total balance is then F1 er1 +r2 + er1 +r2 (F2 − F1 ) =
F2 er1 +r2 . Deposit this on the money market account.
And so on, up to day T when the total balance is FT er1 +···+rT = Xer1 +···+rT .
In this way we have a strategy which is equivalent to a contract where we pay F0
day zero, and receive the value FT er1 +···+rT = Xer1 +···+rT day T . This proves d).
Since relation d) is true for any random variable X whose outcome is known
day T , we may replace X by Xe−R(0,T ) in that relation. This proves e).
It is now easy to prove b). The fact that P (T ) [1] = ZT is simply the definition
of ZT . The relation G(T ) [1] = 1 now follows from c) with X = 1. In order to
(T )
prove that F0 [1] = 1, note that by the definition of money market account, the
price needed to be paid day zero in order to receive eR(0,T ) day T is 1; hence
(T )
1 = P (T ) [eR(0,T ) ]. The relation F0 [1] = 1 now follows from d) with X = 1.
Finally, to prove a), note that if we buy a contracts which cost P (T ) [X] day
zero and gives the payoff X day T , and b contracts that gives payoff Y , then we
have a portfolio that costs aP (T ) [X] + bP (T ) [Y ] day zero and yields the payoff
aX + bY day T ; hence P (T ) [aX + bY ] = aP (T ) [X] + bP (T ) [Y ]. The other two
relations now follow immediately employing c) and d). This completes the proof.

Comparison of Forward- and Futures Prices

Assume first that the short interest rates ri are deterministic, i.e., that their
values are known already day zero. This means that eR(0,T ) is a constant (non-
random,) so

1 = P (T ) [eR(0,T ) ] = P (T ) [1]eR(0,T ) = ZT eR(0,T ) ,

so
ZT = e−R(0,T ) .

Hence

(T ) (T ) (T )
ZT G(T ) [X] = P (T ) [X] = F0 [Xe−R(0,T ) ] = F0 [X]e−R(0,T ) = F0 [X]ZT

and hence
(T )
G(T ) [X] = F0 [X].

We write this down as a corollary:

3
Lecture Note 1

Corollary
If interest rates are deterministic, the forward price and the futures price
(T )
coincide: G(T ) [X] = F0 [X]

The equality of forward- and futures prices does not in general hold if interest
rates are random, though. To see this, we show as an example that if eR(0,T ) is
(T )
random, then F0 [eR(0,T ) ] > G(T ) [eR(0,T ) ].
Indeed, note that the function y = x1 is convex for x > 0. This implies that
1
its graph lies over its tangent. Let y = m + k(x − m) be the tangent line through
1 1 1
the point (m, m ). Then x ≥ m + k(x − m) with equality only for x = m (we
consider only positive values of x.) Now use this with x = e−R(0,T ) and m = ZT .
We then have
eR(0,T ) ≥ ZT−1 + k(e−R(0,T ) − ZT )

where the equality holds only for one particular value of R(0, T ). In the absence
of arbitrage (we will come back to this in Lecture Note 4), the futures price of the
value of the left hand side is greater than the futures price of the value of the right
hand side, i.e.,
(T ) (T )
F0 [eR(0,T ) ] > ZT−1 + k(F0 [e−R(0,T ) ] − ZT )

(T )
But F0 [e−R(0,T ) ] = P (T ) [1] = ZT , so the parenthesis following k is equal to zero,
hence
(T )
F0 [eR(0,T ) ] > ZT−1

On the other hand,

ZT G(T ) [eR(0,T ) ] = P (T ) [eR(0,T ) ] = 1

(T )
so G(T ) [eR(0,T ) ] = ZT−1 , and it follows that F0 [eR(0,T ) ] > G(T ) [eR(0,T ) ].

In general, if X is positively correlated with the interest rate, then the futures
price tends to be higher than the forward price, and vice versa.

Spot prices, storage cost and dividends

Consider a forward contract on some asset to be delivered at a future time
T . We have talked about the forward price, i.e., the price paid at the time of
delivery for the contract, and the present price, by which we mean the price paid
for the contract today, but where the underlying asset is still delivered at T . This
should not be confused with the spot price today of the underlying asset. The
present price should equal the spot price under the condition that the asset is
an investment asset , and that there are no storage costs or dividends or other
benefits of holding the asset. As an example: consider a forward contract on a
share of a stock to be delivered at time T . Let r be the interest rate (so that
ZT = e−rT ) and S0 the spot price of the share. Since 1 euro today is equivalent
to erT euro at time T , the forward price should then be G = S0 erT . But only
if there is no dividend of the share between now and T , for if there is, then one

4
Lecture Note 1

could make an arbitrage by buying the share today, borrow for the cost and take
a short position on a forward contract. There is then no net payment today, and
none at T (deliver the share, collect the delivery price G of the forward and pay
the S0 erT = G for the loan.) But it would give the trader the dividend of the
share for free, for this dividend goes to the holder of the share, not the holder of
the forward. Likewise, the holder of the share might have the possibility of taking
part in the annual meeting of the company, so there might be a convenience yield.

5
Lecture Note 2

Note 2: Forwards, FRA:s and Swaps

Forward prices
In many cases the theorem of Lecture Note 1 can be used to calculate forward
prices. As we will see later, in order to calculate option prices, it is essential to
first calculate the forward price of the underlying asset.

Example 1.
Consider a share of a stock which costs S0 today, and which gives a known
dividend amount d in t years, and whose (random) spot price at time T > t is ST .
Assume that there are no other dividends or other convenience yield during the
time up to T . What is the forward price G on this stock for delivery at time T ?
Assume that we buy the stock today, and sell it at time T . The present value
of the dividend is Zt d and the present value of the income ST at time T is ZT G.
Hence we have the relation
S 0 = Zt d + ZT G
from which we can solve for G

Example 2.
Consider a share of a stock which costs S0 now, and which gives a known
dividend yield d St i t years, where St is the spot price immediately before the
dividend is payed out. Let the (random) spot price at time T > t be ST . Assume
that there are no other dividends or other convenience yield during the time up
to T . What is the forward price G on this stock for delivery at time T ?
Consider the strategy of buying the stock now, and sell it at time t immedi-
ately before the dividend is paid out. With the notation of Lecture Note 1, we
have the relation
P (t) (St ) = S0 (1)
Consider now the strategy of buying the stock now, cash the dividend at time
t, and eventually sell the stock at time T . The present value of the dividend is
d P (t) (St ) and the present value of the income ST at time T is ZT G. Hence we
have the relation
S0 = d P (t) (St ) + ZT G
If we combine with (1) we get

(1 − d)S0 = ZT G

from which we can solve for G

Example 3.
With the same setting as in example 2, assume that there are dividend yield
payments at several points in time t1 , . . . , tn , each time with the amount d Stj . As
in the above example, we can buy the stock today and sell it just before the first
dividend is paid out, so the relation

P (t1 ) (St1 ) = S0 (2)

6
Lecture Note 2

holds. For any k = 2, 3, . . . , n we can buy the stock at time tk−1 immediately before
the payment of the dividend, collect the dividend, and sell the stock immediately
before the dividend is paid out at time tk . The price of this strategy today is zero,
so we have

0 = −P (tk−1 ) (Stk−1 ) + d P (tk−1 ) (Stk−1 ) + P (tk ) (Stk ), i.e.,

P (tk ) (Stk ) = (1 − d)P (tk−1 ) (Stk−1 )

and repeated use of this relation gives

P (T ) (ST ) = (1 − d)n P (t1 ) (St1 ) = (1 − d)n S0

where we have used (2) to obtain the last equality. Hence, by the theorem of
Lecture Note 1, we have the relation

ZT G = (1 − d)n S0 (3)

Example 4.
We now consider the setting in example 3, but with a continuous dividend
yield ρ, i.e., for any small interval in time (t, t + δt) we get the dividend ρSt δt.
If we divide the time interval (0, T ) into a large number n of intervals of length
δt = T /n, we see from (3) that

ZT G = (1 − ρ δt)n S0

and when n → ∞ we get

ZT G = e−ρT S0

Example 5.
Assume we want to buy a foreign currency in t years time at an exchange
rate, the forward rate, agreed upon today. Assume that the interest on the foreign
currency is ρ and the domestic rate is r per year. Let X0 be the exchange rate
now (one unit of foreign currency costs X0 in domestic currency,) and let Xt be
the (random) exchange rate as of time t. Let G be the forward exchange rate.
Consider now the strategy: buy one unit of foreign currency today, buy foreign
zero coupon bonds for the amount, so that we have eρt worth of bonds at time
t when we sell the bonds. Since the exchange rate at that time is Xt , we get
Xt eρt in domestic currency. Since the price we have paid today is X0 we have the
relation
¡ ¢
X0 = P (t) Xt eρt = e−rt G eρt = G e(ρ−r)t

7
Lecture Note 2

Forward Rate Agreements

A forward rate agreement is a forward contract where the parties agree that
a certain interest rate will be applied to a certain principal during a future time
period. Let us say that one party is to borrow an amount L at time t and later
pay back the amount Lef (T −t) at time T > t. The cash flow for this party is thus
L at time t and −Lef (T −t) at time T . Since this contract costs nothing now, we
have the relation
0 = Zt L − ZT Lef (T −t) .

From this we can solve for f . The interest rate f is the forward rate from t to T .

Plain Vanilla Interest Rate Swap

The simplest form of an interest swap is where one party, say A, pays party
B:
the floating interest on a principal L1 from time t0 to t1 at time t1
the floating interest on a principal L2 from time t1 to t2 at time t2
the floating interest on a principal L3 from time t2 to t3 at time t3
···
the floating interest on a principal Ln from time tn−1 to tn at time tn .

The floating rate between tj and tk is the zero coupon rate that prevails between
these two points in time. The amount that A pays at time tk is thus
Lk · (1/Z(tk−1 , tk ) − 1), where Z(tk−1 , tk ) of course is the price of the zero coupon
bond at time tk−1 that matures at tk . Note that this price is unknown today but
known at time tk−1 . The total amount that B will receive, and A will pay is thus
random.
On the other hand, party B pays A a fixed amount c at each of the times
t1 , . . . , tn . The question is what c ought to be in order to make this deal “fair”.

Consider the following strategy: Do nothing today, but at time tk−1 buy
1/Z(tk−1 , tk ) zero coupon bonds that mature at time tk . This gives a cash flow of
−1 at time tk−1 and 1/Z(tk−1 , tk ) at time tk The price today of this cash flow is
zero, so

0 = −Ztk−1 + P (tk ) (1/Z(tk−1 , tk )) i.e.,

¡ ¢
P (tk ) 1/Z(tk−1 , tk ) = Ztk−1

We can now calculate the present value Pk of the payment of A to B at time tk :

Pk = Lk · P (tk ) (1/Z(tk−1 , tk ) − 1) = Lk (Ztk−1 − Ztk )

So the total present value of A:s payments to B is

n
X
PAB = Lk (Ztk−1 − Ztk )
1

8
Lecture Note 2

The present value of B:s payments to A is, on the other hand,

n
X
PBA = c Zt k
1

so we can calculate the fair value of c by solving the equation we get by setting
PBA = PAB .

9
Lecture Note 3

Note 3: Optimal Hedge Ratio

A company that knows that it is due to buy an asset in the future can hedge by
taking a long futures position on the asset. Similarly, a firm that is going to sell
may take a short hedge. But it may happen that there is no futures contract on
the market for the exact product or delivery date. For instance, the firm might
want to buy petrol or diesel, whereas the closest futures contract is on crude oil,
or the delivery date of the futures contract is a month later than the date of the
hedge. In this case one might want to use several futures on different assets, or
delivery times, to hedge. Let S be the price of the asset to be hedged at the date
t of delivery, and Ft1 , . . . , Ftn the futures prices at date t of the contracts that are
being used to hedge. All these prices are of course random as seen from today,
but we assume that there are enough price data so that it is possible to estimate
their variances and covariances.
Assume that for each unit of volume of S we use futures contracts correspond-
ing to βi units of volume for contract F i . The difference between the spot Pn price S
and the total futures price at the date of the end of the hedge is e = S − i=1 βi Fti ,
or
Xn
S= βi Fti + e
i=1
Pn
ThePtotal price paid for the asset including the hedge is S − i=1 βi (Fti − F0i ) =
n
e + i=1 βi F0i , where only e is random, i.e., unknown to us today. The task is to
choose βi such that the variance of the residual e is minimised.

Lemma
If we choose the coefficients βi such that Cov(Fti , e) = 0 for i = 1, . . . , n, then
the variance Var(e) is minimised.

Proof
Assume that we have chosen βi such that Cov(Fti , e) = 0 for i = 1, . . . , n, and
consider any other choise of coefficients:
n
X
S= γi Fti + f
i=1

For notational convenience, let δi = βi − γi . We can write the residual f as

n
X n
X n
X
f =S− γi Fti = βi Fti + e − γi Fti
i=1 i=1 i=1
n
X
= δi Fti + e
i=1
Pn
Note that since Cov(Fti , e) = 0 it holds that Cov( i=1 δi Fti , e) = 0, hence
³X
n ´
Var(f ) = Var δi Fti + Var(e) ≥ Var(e)
i=1

10
Lecture Note 3

Q.E.D.

Theorem
The set of coefficients βi that minimises the variance Var(e) of the residual is
the solution to the system
n
X
Cov(Ftj , Fti )βi = Cov(Ftj , S) j = 1, . . . , n.
i=1

Proof
The condition that Cov(Ftj , e) = 0 means that
n
X
0= Cov(Ftj , S − βi Fti )
i=1
Xn
= Cov(Ftj , S) − βi Cov(Ftj , Fti )
i=1

from which the theorem follows. Q.E.D.

If we regard S and Fti as any random variables, then the coefficients βi are called
the regression coefficients of S onto Ft1 , . . . , Ftn ; in the context here they are the
optimal hedge ratios when the futures Ft1 , . . . , Ftn are used to hedge S: for each
unit of volume of S we should use futures Fti corresponding to βi units of volume
in the hedge.
Let us consider the case n = 1. In this case we have that

Cov(S, Ft )
β=
Var(Ft )

Using the lemma, we have the following relation of variances:

2 Cov2 (S, Ft )
Var(S) = β Var(Ft ) + Var(e) = + Var(e)
Var(Ft )
= ρ2 (S, Ft ) Var(S) + Var(e)
p
where ρ(S, Ft ) = Cov(S, Ft )/ Var(S) Var(Ft ) is the correlation coefficient be-
tween S and Ft . ¡ ¢
Solving for Var(e) we get the plesant relation Var(e) = Var(S) 1 − ρ2 (S, Ft )
p
which means that the standard deviation of the hedged position is 1 − ρ2 (S, Ft )
times that of the unhedged position.

11
Lecture Note 4

Note 4: Conditions for No Arbitrage

Let us consider a two period market model of contracts where one party, the “long”
holder, pays a sum P ,—the present price—today when the contract is drawn up,
and the other part delivers some value X at some later date t. A contract is
defined by the random payoff X and the present price P ; we describe it by the
pair (P, X). The total cash flow for the long holder at time t is thus X, which
is random, whereas the cash flow today is −P and is deterministic. There is no
other cash flow in relation to entering the contract.
The market consists of a set of contracts, and we assume that we can com-
pose arbitrary portfolios of contracts, i.e., if we have contracts 1, . . . , n with payoffs
X1 , . . . , Xn and present prices P1 , . . . , Pn , then we can compose a portfolio con-
sisting of λi units of contract i, i = 1, . . . , n where λ1 , . . . , λn are arbitrary real
numbers. I.e., we assume that we can take a short position in any contract, and
ignore anyPdivisibility problems. The total Pnpresent price of such a portfolio is
n
of course 1 λi Pi and the total payoff is 1 λi Xi . We call also such portfolios
“contracts”, so the set of contracts, defined as pairs (P, X), constitute a linear
space.
We assume that the market is arbitrage free in the sense that the law of one
price prevails (Lecture Note 1,) but also in the sense that
1. If X ≥ 0, then P (X) ≥ 0.
2. If X = 0 almost surely (abbr. “a.s.”, i.e., with probability 1) then P (X) = 0.
3. If X ≥ 0 and P (X) = 0, then X = 0 a.s.

These conditions avoid obvious arbitrage opportunities.

The operator P (X) behaves very much like an expectations operator for a
probability measure. We will now show that if the underlying sample space is
finite, then there is a random variable U such that U > 0, and such that for any
payoff X, P (X) = E [XU ].
It is obvious that the opposite is true: If there is a random variable Q with
the above properties, then the market satisfies the no arbitrage conditions 1) – 3).

Theorem
If the sample space is finite and the market is arbitrage free in the sense given
above, then there exists a random variable U such that U > 0, and for any
payoff X it holds that P (X) = E [XU ].

Proof
Let Ω = {ω1 , . . . , ωn , . . . , ωN } be the sample space and without loss of gen-
erality we may assume that ωk have positive probabilities pk for k ≤ n and zero
probability for k > n. ¡ ¢
We associate with each contract (P, X) the vector − P, X(ω1 ), . . . , X(ωn )
in Rn+1 i.e., we ignore X:s values on events with probability zero. The set of
such vectors constitute a linear subspace V of Rn+1 . We now prove that there is
a vector q̄ which is orthogonal to V all of whose coordinates are positive.
Let K be the subset of Rn+1 : K = {ū = (u0 , . . . , un ) ∈ Rn+1 | u0 +· · ·+un =
1 and ui ≥ 0 for all i}. Obviously K and V have no vector in common; indeed, it

12
Lecture Note 4

is easy to see that any such common vector would represent an arbitrage. Now let
q̄ be the vector of shortest Euclidean length such that q̄ = ū − v̄ for some vectors
ū and v̄ in K and V respectively. The fact that such a vector exists needs to be
proved, however, we will skip that proof. We write q̄ = ū∗ − v̄ ∗ where ū∗ ∈ K and
v̄ ∗ ∈ V.
Now note that for any t ∈ [0, 1] and any ū ∈ K, v̄ ∈ V , the vector tū + (1 −
t)ū ∈ K and tv̄ + (1 − t)v̄ ∗ ∈ V, hence |(tū + (1 − t)ū∗ ) − (tv̄ + (1 − t)v̄ ∗ )| as a
∗

function of t on [0, 1] has a minimum at t = 0, by definition of ū∗ and v̄ ∗ , i.e.,

|t(ū − v̄) + (1 − t)q̄|2 = t2 |ū − v̄|2 + 2t(1 − t)(ū − v̄) · q̄ + (1 − t)2 |q̄|2 has minimum
at t = 0 which implies that the derivative w.r.t. t at t = 0 is ≥ 0. This gives
(ū − v̄) · q̄ − |q̄|2 ≥ 0 or, equivalently, ū · q̄ − |q̄|2 ≥ v̄ · q̄ for all v̄ ∈ V and ū ∈ K.
But since V is a linear space, it follows that we must have v̄ · q̄ = 0 for all v̄ ∈ V.
It remains to prove that q̄ has strictly positive entries. But we have ū · q̄ − |q̄|2 ≥ 0
for all ū ∈ K, in particular we can take ū = (1, 0, . . . , 0) which shows that the first
entry of q̄ is > 0 and so on.
Hence, we have found positive numbers q0 , . . . , qn such that

−q0 P + q1 X(ω1 ) + · · · + qn X(ωn ) = 0

for all contracts, and by scaling the qi :s we can arrange that q0 is equal to one,
hence
Xn
P = qi X(ωi ) (1)
1

We now define U (ωi ) = pqii for i = 1, . . . , n and U (ωi ) = arbitrarily positive for
i = n + 1, . . . , N . U is thus a positive random variable, and it is easy to verify
that it satisfies the conditions of the theorem. Q.E.D.

The No Arbitrage Assumption

It is easy to see that also in the case of an infinite sample space, it is a sufficient
condition for the absence of arbitrage that there exists a random variable U
as in the theorem. In the sequel we will always assume the existence of such
a random variable U .

Note that by definition, the zero coupon price Zt = P (1) = E [U ]. Define the
random variable Q by Q = Zt−1 U , then Q > 0, E [Q] = 1 and

P (X) = Zt E [XQ]

which shows that the expectation is equal to the forward price:

G(X) = E [XQ].

13
Lecture Note 5

Note 5: Pricing European Derivatives

Black’s Model
Let us compute the price of a European derivative on some underlying asset
with value X at maturity t. We assume that there already is a forward contract
on the market, or that we can compute the forward price, and denote the current
forward price for delivery at t by G. Black’s model assumes that the value X has
a log-normal probability distribution:
√
X = Aeσ tz
where z ∈ N (0, 1) (1)

A European derivative on X is a contract on a payoff of some function f (X) of X.

In accordance with Lecture Note 4 we assume the existence of a random variable
Q such that Q > 0, E [Q] = 1 and such that forward prices are set according to
¡ ¢
G f (X) = E [f (X)Q]

In this formula, Q can be replaced by E [Q | z] which is a function of z, so in

the current context, we can regard Q to be a function of z. A natural choice is
a log-normal specification: Q = ea+bz for some constants a and b. The condition
E [Q] = 1 imposes the restriction a = − 12 b2 . For notational convenience we choose
√
b = −λ t, i.e.,
1 2
√
Q = e− 2 λ t−λ t z

Now we can express the forward value of the underlying asset as

√ 1 2
√ 1 2
G(X) = E [XQ] = E [Aeσ tz
e− 2 λ t−λ t z
] = Ae 2 σ t−σλt

1 2
Hence A = G(X)eσλt− 2 σ t
which we substitute into (1):
1 2
√
X = G(X)eσλt− 2 σ t+σ t z
(2)

The forward price of any derivative of X is thus

¡ ¢ £
G f (X) = E f (X)Q]
¡ 1 2
√ ¢ 1 2
√ ¤
= E [f G(X)eσλt− 2 σ t+σ t z e− 2 λ t−λ t z
Z ∞
1 ¡ 1 2
√ ¢ 1 2
√ 1 2
=√ f G(X)eσλt− 2 σ t+σ t x e− 2 λ t−λ t x e− 2 x dx
2π −∞
Z ∞
1 ¡ 1 2
√ ¢ 1
√ 2
=√ f G(X)eσλt− 2 σ t+σ t x e− 2 (x+λ t) dx
2π −∞
£ √
= substitute x + λ t = y]
Z ∞
1 ¡ 1 2
√ ¢ 1 2
=√ f G(X)e− 2 σ t+σ t y e− 2 y dy
2π −∞
¡ 1 2
√ ¢¤
= E [f G(X)e− 2 σ t+σ t w where w ∈ N (0, 1)

14
Lecture Note 5

Finally, if G denotes the forward price of the underlying asset, the present
price of the derivative is:
¡ 1 2
√ ¢¤
p = Zt E [f Ge− 2 σ t+σ t w where w ∈ N (0, 1) (3)

This is Blacks pricing formula.

The Black-Scholes Pricing Formula

Let us consider the case when the underlying asset is an investment asset
with no dividends or convenience yield, for example, it might be a share of a stock
which pays no dividend before maturity of the contract. In this case there is a
simple relationship between the forward price G and the current spot price S0 of
the underlying asset. Indeed, G = ert S0 where r is the t-year zero coupon interest
rate. To see this, note that S0 = P (t) = Zt G, i.e., S0 = e−rt G, see Lecture Note
2, example 1 with d = 0. Hence the present price of the derivative can be written
¡ 1 2
√ ¢¤
p = e−rt E [f S0 ert− 2 σ t+σ t w where w ∈ N (0, 1)

This is the Black-Scholes pricing formula.

Put and Call Options

A call option with strike price K is specified by f (X) = max[X − K, 0], and
a put option by f (X) = max[K − X, 0]. In these cases the expectation in Black’s
formula can be evaluated, and the result is that the call option price c and put
option price p are
¡ ¢
c = Zt GΦ(d1 ) − KΦ(d2 )
¡ ¢
p = Zt KΦ(−d2 ) − GΦ(−d1 )
where
ln(G/K) 1 √
d1 = √ + σ t
σ t 2
√
d2 = d1 − σ t
and Φ is the cumulative normal distribution

The Interpretation of σ and λ

The parameter σ is called the volatility of the asset X. It is the standard
deviation of the logarithm of X divided by the square-root of the time to maturity.
If the value of X is some constant times the product of identically distributed
independent random variables whose outcome occur every day (or hour, or any
short time interval,) then ln X will be approximately normally distributed. Since
the variance is proportional to the number of days (or whatever interval used,)
the variance is proportional to the time interval until maturity. So the standard
deviation is proportional to the square-root of the time interval to maturity.

15
Lecture Note 5

If we compute the expected value in (2), we get

E [X] = Geλσt

which means that the expected rate of return on a forward contract on X is λσ.
The expected rate of return is thus proportional to the standard deviation σ, and
λ is called the market price of risk.
The situation when λ > 0 is called backwardation, and when λ < 0 the
situation is known as contango.

16
Lecture Note 6

Note 6: Yield and Duration

Consider at time t = 0 a zero coupon bond which matures at time t = D with a
face value of PD . The present value of this bond is of course P0 (y0 ) = PD e−y0 D
where y0 is the relevant zero coupon interest rate. Note that
1 0
D=− P (y0 ) (1)
P0 0
P0 (y) = PD e−yD (2)
−y(D−t)
Pt (y) = PD e (3)

where Pt (y) is the value at time t, if the zero coupon interest rate at that time
equals y, which may differ from y0 .
The idea of the concepts yield (y) and duration (D) is to define a number D
for a coupon paying bond, or a portfolio of bonds, such that the relations (2) and
(3) still hold approximately. Hence, let P0 be the present value of a bond, or a
portfolio of bonds:
Xn
P0 = Zti ci
i=1

where Zs is the price today of a zero coupon bond maturing at time s and ci is
the payment received at time ti . Now define the yield y0 by the relation
n
X
P0 = e−y0 ti ci = P0 (y0 ) (4)
i=1

and the duration D by

n
1 0 1 X −y0 ti
D = − P0 (y0 ) = ti e ci (5)
P0 P0 i=1

Now define P̂D (y) = P0 (y)eyD [see (4).] We then have

0
P̂D (y0 ) = P00 (y0 )ey0 D + DP0 (y0 )ey0 D = 0,

hence we can consider P̂D (y) to be constant regarded as a function of y, to a

first order approximation. Henceforth we will treat P̂D = P̂D (y0 ) = P0 ey0 D as
constant. We now have

P0 (y) = P̂D e−yD (20 )

Pt (y) = P̂D e−y(D−t) (30 )

Here (20 ) follows from the definition of P̂D and (30 ) holds if the yield is y at time
t, as is seen as follows:
n
X n
X
−y(ti −t) yt
Pt (y) = e ci = e e−yti ci = eyt P0 (y) = P̂D e−y(D−t)
i=1 i=1

17
Lecture Note 6

Summary
A bond, or a portfolio of bonds, has a yield y0 , defined by (4), and a duration
D, defined by (5). At any time t before any coupon or other payments have been
paid out, the value Pt (y) of the asset is approximately equal to

Pt (y) = P̂D e−y(D−t) where P̂D = P0 ey0 D

and y is the yield prevailing at that time. In particular, the value of the asset at
time t = D is P̂D = P0 ey0 D , and hence independent of any changes in the yield to
a first order approximation.

Example
Consider a portfolio of bonds that gives the payment 1’000 after one year,
1’000 after two years and 2’000 after three years. Assume that Z1 = 0.945, Z2 =
0.890 and Z3 = 0.830. The present value of the portfolio is then

P0 = 0.945 · 1’000 + 0.890 · 1’000 + 0.830 · 2’000 = 3’495

The yield y0 is obtained by solving for y0 in the relation

3’495 = P0 = e−y0 1’000 + e−2y0 1’000 + e−3y0 2’000

which gives y0 = 0.06055. Now

−P00 (y0 ) = e−y0 1’000 + 2 e−2y0 1’000 + 3 e−3y0 2’000 = 7’716.5

so the duration is D = 7’716.5 years = 2.208 years. The value of the portfolio
3’495
at time t, if the yield at that time is y and t is less than one year, can thus be
approximated by
Pt (y) = 3’995 e−y(2.208−t)

(where 3’995 = P0 ey0 D .)

Forward Yield and Forward Duration

Consider a bond, or any interest rate security, that after time t gives the
holder deterministic payments c1 , c2 , . . . , cn at times t1 < t2 < · · · < tn (where
t1 > t.) Today is time 0 < t and we want to compute the forward price of the
security to be delivered at time t. Note that the security may give the holder
payments also before time t, but this is of no concern to us, since we only are
interested in the forward value of the security.
Let Pt be the (random) value of the security at time t and consider the
strategy of buying the security at time t for this price and then cash the payments
c1 , . . . , cn . The pricing methods of Lecture Note 2 shows that the forward price
G(t) of the value Pt is given by
n
X
(t)
Zt G = Zti ci
i=1

18
Lecture Note 6

The forward yield yF is defined by

n
X
(t)
G = ci e−yF (ti −t) (6)
i=1

When we arrive at time t it may be that the prevailing yield y at that time
is not the same as yF . Let y be the prevailing yield at time t. Then the value of
the security at that time is
n
X
Pt = Pt (y) = ci e−y(ti −t)
i=1

In particular, we see that Pt (yF ) = G(t) . We define the forward duration DF of

the portfolio for time t as
n
Pt0 (yF ) 1 X
DF = − = (t) ci (ti − t)e−yF (ti −t) . (7)
Pt (yF ) G i=1

Now define PF by PF (y) = Pt (y)eDF y . The same calculation as in the previous

section shows that PF is independent of y, to a first order approximation, so we
will regard PF = PF (yF ) = G(t) eDF yF as a constant. So we have

Pt = PF e−DF y

Summary
An interest rate security that after time t gives the deterministic payments
c1 , c2 , . . . , cn at times t1 < t2 < · · · < tn has a forward yield yF , defined by (6),
and a forward duration DF defined by (7). The value of the asset as of time t,
0 < t < t1 is then, to a first order approximation

Pt = PF e−DF y where PF = G(t) eDF yF

and y is the yield prevailing at that time.

Example
In the previous example, assume that Z1.5 = 0.914. The forward price G1.5
of the portfolio for delivery in one and a half years is obtained from the relation
(note that the first payment of 1’000 has already been paid out, and hence does
not contribute to the forward value:)

0.914 G1.5 = 0.890 · 1’000 + 0.830 · 2’000 ⇒ G1.5 = 2’790

The forward yield yF is obtained from

2’790 = G1.5 = e−0.5yF 1’000 + e−1.5yF 2’000 ⇒ yF = 0.06258

19
Lecture Note 6

Now
−G01,5 (yF ) = 0.5 e−0.5yF 1’000 + 1.5 e−1.5yF 2’000 = 3’215.8

so the forward duration is DF = 3’215.8 years = 1.153 years.

2’790

Black’s Model for Bond Options

Consider again the bond or security in the section “Forward Yield and Forward
Duration” above. In this section we want to price a European option on the value
Pt , a bond option.
First we specify a random behaviour of y, the yield that will prevail for the
underlying security at time t. We will assume that as seen from today, y is a
normally distributed random variable:
√
y = α − σ t z, where z ∈ N(0, 1)

The minus sign is there for notational convenience later on; note that z and −z
have the same distribution. Hence, the present value Pt at time t of the asset is
√ √
Pt = PF e−DF (α−σ t z)
= A eDF σ tz

where A = PF e−DF α .
Note that this is exactly the same specification as (1) in Lecture Note 5 with
σ replaced by DF σ. We √ can thus use Black’s formula for European options of
Lecture Note 5, where σ t in that formula stands for the product of the standard
deviation of the forward yield and the forward duration of the underlying asset.

Portfolio Immunising
Assume that we have portfolio P of bonds whose duration is DP . The duration
reflects its sensitivity to changes in the yield, so we might well want to change
that without making any further investments. One way to do this is to add a bond
futures to the portfolio (or possibly short such a futures.)
Consider again the security discussed under “Forward Yield and Forward Du-
ration” above. We will look at a futures contract on the value Pt . Using the no-
tation of Lecture Note 1, the futures price is F0 = F0 (Pt ) = F0 (PF e−DF y ), where
the yield y at time t is random as seen from today. Assume now that the present
yield y0 of the portfolio P changes to y0 + δy. The question arises what impact this
has on y. If y = (current yield) + (random variable independent of current yield),
then y should simply be replaced by y + δy. The new futures price should then be

F0 (PF e−DF (y+δy) ) = e−DF δy F0 (PF e−DF y ) = e−DF δy F0

The “marking-to-market” is hence (e−DF δy − 1)F0 .

Assume now that we add such a futures contract to our initial portfolio P .
We let P = P (y0 ) denote its original value, and P (y0 + δy) its value after the yield
has shifted by δy. The value of the total portfolio Ptot before the shift is equal to
P , since the value of the futures contract is zero. However, after the yield shift,

20
Lecture Note 6

the marking-to-market is added to the portfolio, so after the shift the value of the
total portfolio is

Ptot (y0 + δy) = P (y0 + δy) + (e−DF δy − 1)F0

If we differentiate this w.r.t. δy we get for δy = 0

0
−Ptot (y0 ) = −P 0 (y0 ) + DF F0 = DP P + DF F0

The duration of the total portfolio is thus

1 0 F0
Dtot = − Ptot (y0 ) = DP + DF
P P

If we take N futures contracts (where negative N corresponds to going short the

futures,) we get
F0
Dtot = DP + N DF
P

21
Lecture Note 7

Note 7: Risk Adjusted Probability Measures

In Lecture Note 4 we introduced the notion that a forward price can be expressed as
an expectation: G(t) (X) = E [XQ] for some random variable Q > with E [Q] = 1.
In probability theory, what we have done is called a change of probability measure.
If Q is as above, then it is convenient to introduce a notation for the expected
value E [XQ], for example
E∗ [X] = E [XQ]

and say that E∗ [X] is the expected value of X with respect to a new probability
measure Pr∗ , where the probability of an event A with this new probability measure
is defined by
Pr∗ (A) = E∗ [IA ] = E [IA Q]

where IA is the indicator function of A, i.e., IA (ω) is equal to one if ω ∈ A and

zero otherwise. It is easy to check that Kolmogorov’s axioms for probabilities are
satisfied for the probabilities Pr∗ (A), and also that the expextation E∗ [·] satisfies
the usual rules for an expectation:
E∗ [X] ≥ 0 if X ≥ 0,
E∗ [1] = 1,
E∗ [λ1 X1 + λ2 X2 ] = λ1 E∗ [X1 ] + λ2 E∗ [X2 ],
if λ1 , λ2 are real numbers and X1 , X2 random variables.
The random variable Q is called the Radon-Nikodym derivative of the new proba-
bility measure w.r.t. the natural measure. When U is defined as in Lecture Note
(t)
4, we have G0 (X) = E [XQ] and we call the new probability measure the forward
measure with maturity date t, and use the notation E(t) for the expectation, i.e.,

(t)
G0 (X) = E(t) [X]

Choise of Numeraire
The forward probability measure introduced above is a very important exam-
ple of a risk adjusted probability measure, but one can also use other numeraires.
Let N be some random value whose outcome becomes known at some date t, and
assume that N > 0. For instance, N may be the value of one barrel of crude oil
at time t. Now define the random variable QN by

1
QN = (t)
UN
P0 (N )

where U is defined as in Lecture Note 4. Then QN > 0 and E [QN ] = 1 (verification

left to the reader) and we have for any random variable X whose outcome is known
at time t
(t) (t) £X ¤
P0 (X) = P0 (N ) E QN .
N

22
Lecture Note 7

The verification is left as an exercise for the reader. If we denote by EN [·] the
expectation associated with QN , then

(t) (t) £X ¤
P0 (X) = P0 (N ) EN (1)
N

Here N is the numeraire. For example, if X is the price of jet fuel at time t and
the numeraire N is the price of crude oil at time t, the idea could be that we want
to calculate the present price for jet fuel to be delivered at time t, but rather than
creating a model for the price of jet fuel, we might find it easier to come up with a
reasonable model for how the market perceives the £ ratio
¤ X/N of jet fuel to crude
N X
oil. We can then from that model compute E N , whereas the present price
(t)
P0 (N ) can be directly observed on the market (since there is a futures market
for crude oil.)
If we choose N = one unit of some currency , i.e., the numeraire asset is a
zero coupon bond, then we have

(t) (t)
P0 (X) = P0 (1) EN [X] = Zt EN [X]

(t)
i.e., EN = G0 = E(t) , the forward mesaure with maturity date t.

Forward Measures for Different Maturities

The forward expectation E(t) depends on the time of maturity t; i.e., in gen-
eral, if t < T , then E(t) (X) 6= E(T ) (X) However, if interest rates are deterministic,
then the two coincide; more specifically, we have the following theorem: let Z(t, T )
be the price at time t of a zero coupon bond maturing at T ;

Theorem 1
Assume that t < T and that the zero coupon bond price Z(t, T ) is known at
the present time 0. Then, for any random variable X whose outcome is known
at time t,
E(t) [X] = E(T ) [X]

Proof:
Consider the following strategy: Enter a contract forward contract on X
maturing at t (so G(t) = E(t) [X],) invest the net payment X − G(t) in zero coupon
bonds maturing at T . In this way we have constructed a contract which pays
Z(t, T )−1 (X − G(t) ) at time T and costs 0 today. Hence

0 = ZT E(T ) [Z(t, T )−1 (X − G(t) )] i.e.,

0 = E(T ) [(X − G(t) )] i.e.,
E(T ) [X] = G(t) = E(t) [X]

which proves the theorem.

23
Lecture Note 7

Conditional Expectations and Martingales

The outcome of random variables may occur at different points in time. For
instance, if I toss a coin, the outcome X, which is heads or tail, is random before
the coin is actually tossed, but at the time it has been tossed, the outcome is
observed.
Let t < T be two points in time, and X a random variable whose outcome is
observed at time T , and Y a random variable whose outcome is observed at the
earlier time t. Assume that for any event A whose outcome can be observed at
time t, the relation
E [IA X] = E [IA Y ] (2)

holds. We then say that Y is the conditional expectation at time t of the variable
X. It is easy to see that the variable Y is essentially uniquely determined by X,
i.e., if Y1 and Y2 both satisfy E [IA X] = E [IA Yi ] for all A, then Y1 = Y2 a.s.
The conditional expectation is denoted Et [X], i.e., Y = Et [X]. Note in par-
ticular the law of iterated expectations
£ ¤
E Et [X] = E [X] (3)

which is obtained by letting A be the entire sample space, so that IA = 1. It is

even true that
£ ¤
Es Et [X] = Es [X] if s < t (30 )

We leave the verification of this relation as an exercise for the reader.

The definition of Et [X] may seem a bit strange, but the interpretation is as
follows: if we at time t form the expectation of X, using all relevant information
available at the time, then Et [X] is the sensible choice. Note that as seen from a
point in time earlier than t, this expectation is random, since we do not have all
the information to be revealed at time t when the expectation is formed.
If “now” is time 0, then the conditional expectation E0 [X] is the same as the
unconditional expectation E [X]. To see this, note that IA for an event whose
outcome is observed now, is either the constant 1 or the constant 0.
Assume now that the time interval from t0 to T is divided into discrete in-
tervals t0 , t1 , . . . , tn = T , and that we have a sequence of random variables Xj ,
j = 0, . . . , n, where the outcome of Xj is observed at time tj . Such a sequence is
a stochastic process. A stochastic process is a martingale if it is true that

Xj = Ej [Xj+1 ] for j = 0, . . . , n − 1. (4)

It follows immediately from (30 ) that in order for a stochastic process to be a

martingale, it suffices that Xj = Ej [Xn ] for j < n.

Martingale Prices
Consider a contract that gives the random payoff X at time T . The forward
(T )
price Gt , 0 ≤ t < T , of this contract at time t is a random variable whose
outcome is determined at time t.

24
Lecture Note 7

Theorem 2
(T )
For any t, 0 ≤ t ≤ T , the forward price Gt equals the conditional expectation

(T ) (T )
Gt = Et [X]

Hence, the forward process has the martingale property (4) w.r.t. the forward
measure, i.e.,
(T ) (T )
Gj = Ej [Gj+1 ]

Proof
Consider the following strategy: Let A be any event whose outcome occurs at
time t. Wait until time t and then, if A has occurred, enter a forward contract on
(T )
X maturing at T with forward price Gt , but if A has not occurred, do nothing.
(T )
We have thus constructed a contract which gives payoff IA (X − Gt ) at time T
which costs 0 today. Hence

(T ) (T )
ZT E(T ) [IA (X − Gt )] = 0 i.e., E(T ) [IA X] = E(T ) [IA Gt ]

Since this is true for any event A whose outcome is known at time t, we have by
the definition of conditional expectation

(T ) (T )
Gt = Et [X]

Q.E.D.

Remark
From this we deduce (c.f. Lecture Note 1) that the present price pt at time t
of the contract yielding X at time T is

(T )
pt = Z(t, T )Et [X]

where Z(t, T ) is the price at time t of a zero coupon bond maturing at time T .
Especially, if the interest rate is deterministic and equal to r, then the present
prices satisfy
pj = e−r ∆t E∗j [pj+1 ] (5)

where pj is the present price as of time tj and ∆t = tj+1 − tj ; E∗ denotes expecta-

tion w.r.t. the forward measure (which is independent of maturity date, according
to Theorem 1.)

25
Lecture Note 8

Note 8: Asset Price Dynamics

In Black’s model, there are only two relevant points in time: the time when the
contract is written, and the time when it matures. In many cases we have to
consider also what happens in between these two points in time.
First some notation. Time is discrete, t = t0 , . . . , tn = T , and we let for
notational simplicity t0 = 0 and all time spells tk − tk−1 = ∆t be equally long;
hence tk = k∆t. Let X be some random value whose outcome becomes known at
time T ; we are interested in evaluating the prices of derivatives of X.

Black-Scholes Dynamics
The Black-Scholes assumption about the random behaviour of X is that
n
X = Aeσ Σ1 zj (1)

where A is some constant, σ a parameter called the volatility, and the zj :s are
stochastically independent normal
√ N(0, ∆t) random variables (∆t is the variance;
the standard deviation is thus ∆t) where the outcome of zj occurs at time tj .
This is the model for the price dynamics under the true probability measure.
But in order to compute prices, we want the dynamics under the forward measure.

Creating the Forward Measure

Let E denote the expectations operator w.r.t. the true probability measure.
Following the ideas of Lecture Note 5, we now specify a random variable Q such
(T ) ¡ ¢
that Q > 0, E [Q] = 1 and such that E [f (X)Q] = G0 f (X)
Following the ideas of Lecture Note 5, a fairly natural guess to find such a
function is to try the specification
n
Q = e µ−λΣ1 zj (2)

for some constants µ and λ. In order for E [Q] = 1 to be satisfied, we must choose

1 1
µ = − ∆t n λ2 = − λ2 T.
2 2

This follows immediately from the fact that

1
E [ey ] = e 2 v (3)

for a normally distributed random variable y ∈ N(0, v), as is easily checked. Our
suggestion thus looks like
2
1
T −λ Σn
Q = e− 2 λ 1 zi (4)

Now we prove

26
Lecture Note 8

Theorem (Girsanov’s theorem)

Let h(z1 , . . . zn ) be a function (of reasonable regularity) of the independent
N(0, ∆t)-variables z1 , . . . , zn . Then

1 n 2 n
E [h(z1 , . . . , zn )e− 2 Σ1 λj ∆t−Σ1 λj zj ] = E [h(w1 − λ1 ∆t, . . . , wn − λn ∆t)]

where also wj are independent N(0, ∆t)-variables.

Proof
This is just a substitution of variables in a multiple integral:

1 2
E [h(z1 , . . . , zn )e− 2 Σλj ∆t−Σλj zj ]
Z Z
1 − 12 Σλ2j ∆t−Σλj xj
= . . . h(x 1 , . . . , x n )e ×
(2π∆t)n/2
1 n 2
× e− 2 ∆t Σ1 xj dx1 . . . dxn
Z Z
1 1 2
= n/2
. . . h(x1 , . . . , xn )e− 2 ∆t Σ(xj +λj ∆t) dx1 . . . dxn
(2π∆t)
= [change of variables: xj + λj ∆t = yj ]
Z Z
1 1 2
= n/2
. . . h(y1 − λ1 ∆t, . . . , yn − λn ∆t)e− 2 ∆t Σyj dy1 . . . dyn
(2π∆t)
= E [h(w1 − λ1 ∆t, . . . , wn − λn ∆t)]

Q.E.D.

Remark
Girsanov’s theorem generally refers to the more general case when ∆t → 0 so
the sum is replaced by an integral (“Itô integral”.)

We can now employ Girsanov’s theorem in our case. In our case λj = λ, and
the Girsanov transformation (4) hence defines a new measure, which will be our
forward measure E(T ) , such that under this measure (1) can be written

n
X = A eσ Σ1 (wj −λ∆t)
n (5)
= B eσ Σ1 wj

for some constant B and where wj are independent N(0, ∆t)-variables under the
(T ) (T ) 1 2
forward measure. Since G0 = E(T ) [X], it follows that B = G0 e− 2 σ T .

27
Lecture Note 8

Binomial Approximation
In almost all cases when the whole price path has to be taken into account, it is
necessary to use some numerical procedure to calculate the price of the derivative.
A useful numerical procedure is to approximate the specification (5) by a√binomial
tree. In (5), we replace wj by a binary variable bj which takes the value −√ ∆t with
probability (under the forward probability measure) 0.5 and the value + ∆t with
probability 0.5. For large values of n, and small ∆t, this is a good approximation—
in fact, as n → ∞ and ∆t → 0 the price of a derivative calculated from the binomial
tree (see next Lecture Note) will converge to the theoretical value it would have
from the specification (5).
The binomial specification is thus
n
X = B eσ Σ1 b j
√
Note that E(T ) [eσbj ] = cosh(σ ∆t). We know from the previous Lecture Note
(T )
that Gk (X) = Ek [X], so
k √
Gk = B eσ Σ1 bj coshn−k (σ ∆t)

Combining this relation with the same for Gk+1 , we get the relation

eσ bk+1 ¡ √
Gk+1 = Gk √ = Gk 1 ± ε) where ε = tanh(σ ∆t)
cosh(σ ∆t)

and the plus and minus sign occur with probability (under the forward probability
measure) 0.5 each.

Remarks
It is worthwhile to note that (5) is identical to (1) except for the constant
A; the volatility σ is the same. The parameter λ that appears in Q has a similar
interpretation as in Lecture Note 5: E [X] = G0 eλσT and is called the market
price of risk.

28
Lecture Note 9

Note 9: The Binomial Model

In the previous Lecture Note we derived the binomial specification of the forward
price dynamics of the asset that will be underlying a derivative. The forward price
refers to a forward contract that matures at the same time T as the derivative,
or later. If we write the dynamics in the usual tree style, we have for the forward
price dynamics

G0 G0 u G0 u2 G0 u3 G0 u4 ···
G0 d G0 du G0 du G0 du3
2
···
G 0 d2 G0 d2 u G0 d2 u2 ···
G0 d3 G0 d3 u ···
G0 d4 ···

In this diagram, each column represents an instant of time, and moving from one
instant to the next (one step right) means that the forward price either moves “up”
or “down”. In this diagram “up” means going one step to the right on the same
line, and “down” going one step to the right on the line below (this configuration
is convenient in a spread sheet.) Under the forward measure the probabilities of
moving up and down are both 21 . The specification of u and d are as follows:
½
u=1+ε √
where ε = tanh(σ ∆t)
d=1−ε

where ∆t is the time spell between columns, and σ is the volatility of the forward
price.

Pricing an American Futures Option

As long as we assume that interest rates are deterministic, the forward prices
in the binomial tree equals the corresponding futures prices. Hence we can use
this tree for calculating the price of an American futures option. In order to do
so, we construct a tree of prices of the futures option. In this tree the prices are
calculated from right to left. If the time for maturity of the option is at T = t4 ,
then we construct the following tree of prices for the futures option:

p p1,0 p2,0 p3,0 p4,0

p1,1 p2,1 p3,1 p4,1
p2,2 p3,2 p4,2
p3,3 p4,3
p4,4

Here the prices in the rightmost column are known, since they represent the value
of the option at maturity when the underlying futures price is that in the corre-
sponding position of the tree of futures prices. Then option prices are calculated
backwards:
¡ ¢
p̂j,k = e−r∆t 0.5 pj+1,k + 0.5 pj+1,k+1 (1)
pj,k = max[p̂j,k , value if exercised] (2)

29
Lecture Note 9

Indeed, p̂j,k = e−r∆t E∗ [pj+1,• ], which is just another way of writing (1), is the
value if the option is not exercised (see (5) of Lecture Note 7,) and the highest
value of this and the value if exercised is the present value. In order to calculate
the value if exercised, we employ the original tree of futures prices, of course. The
end result p is the price of the option today.

American Call Option on a Share of a Stock

If we want to price an American call option on a share of a stock which
pays no dividend before maturity, we need a binomial tree for the current stock
price, not the futures price of the stock. The assumption is still that the forward
price dynamics is as above, and we derive the stock prices from there. This is
easy; according to Lecture Note 2, example 1 with d = 0, the relation between
the current stock price St and the forward price at the same date Gt is St =
Gt e−r(T −t) . The relevant binomial tree for the stock price is thus

S0 S0 er∆t u S0 e2r∆t u2 S0 e3r∆t u3 ···

S0 er∆t d S0 e2r∆t du S0 e3r∆t du2 ···
S0 e2r∆t d2 S0 e3r∆t d2 u ···
S0 e3r∆t d3 ···

The price of the option is then calculated in the same way as above.

Options on Assets Paying Dividends

Known Dividend Amount
Assume we want to price an option on a share of a stock when just before
time i the stock pays a known dividend d. We need to set up a binomial tree for
the current price St of the stock. Let n > i be the period at which time T the
option matures.
For any time j ≤ n we have the relation
½
e−r(i−j)∆t d + e−r(n−j)∆t Gj if j < i
Sj = (3)
e−r(n−j)∆t Gj if i ≤ j ≤ n

This follows as in Lecture Note 2, example 1.

In particular, we have

G0 = S0 ern∆t − d er(n−i)∆t (4)

It is now easy to build the tree of stock prices: First compute G0 from (4), then
construct the tree for Gj , and from these prices, compute Sj from (3).

30
Lecture Note 9

Known Dividend Yield

Now assume instead that at some time between times i−1 and i the stock will
pay a dividend d Si0 where Si0 is the stock price the moment before the dividend is
paid out.
The relation between the stock price Sj and the futures price Gj for times j
later or at time i is as before

Sj = e−r(n−j)∆t Gj j≥i (5)

For times j before i we can compute the stock price in accordance with example
2 of Lecture Note 2:
Sj (1 − d) = e−r(n−j)∆t Gj

from which it follows that

1
Sj = Gj e−r(n−j)∆t j<i (6)
1−d

In particular,
G0 = (1 − d)S0 ern∆t (7)

It is now easy to build the tree of stock prices: First compute G0 from (7), then
construct the tree for Gj , and from these prices, compute Sj from (5) and (6).

31
Lecture Note 10

Note 10: Random Interest Rates: The Futures Measure

We have assumed interest rates to be for the most part deterministic. In order to
study interest rate derivatives, or other situations where interest rates are assumed
to be random, we need a different probability measure than the forward measure.
One reason for this is that when interest rates are random, forward measures for
different maturity dates are not the same. Rather than having zero coupon bonds
as numeraires (see Lecture Note 7) we use the money market account (see Lecture
Note 1) as a numeraire. Hence, let the numeraire N be N = eR(0,t) where we have
used the notation introduced in Lecture Note 1 and 7. Then (see (1) of Lecture
Note 7)
(t) (t) b b
P0 (X) = P0 (eR(0,t) ) E[Xe −R(0,t)
] = E[Xe −R(0,t)
]

b denotes expectation w.r.t. the probability measure associated with the

where E[·]
numeraire N = eR(0,t) . We see from this relation and Lecture Note 1 that

b (t) (t)
E[X] = P0 (XeR(0,t) ) = F0 (X) (1)

so it seems natural to call the probability measure associated with this numeraire
the futures measure. In the litterature it is often called the Equivalent Martingale
Measure (Hull uses the term “traditional risk neutral measure.”)
It is important to note that the forward measures E(T ) differ for different
maturities T when interest rates are random. This is not the case for the futures
measure, though. Indeed, we will now prove two theorems on the futures measure:

Theorem 1
The futures measure as defined above is independent of the date of maturity T
b (n) [X] =
in the following sense: if the outcome of X is known at time tn then E
b (m) [X] if m > n, where we denote by E
E b (k) the futures measure for contracts
maturing at time tk .

Hence there is no need to index the futures measure by maturity date.

Theorem 2
The futures prices {Fj } have the martingale property w.r.t. the futures mea-
sure: Fj = Eb t [Fk ] for j < k. In particular, the futures price F0 is the expected
j
value w.r.t. the futures measure of X, the spot price at delivery.

32
Lecture Note 10

Proof of Theorem 1
Let P be the present price of the value XeR(0,n) to be delivered at time tn .
We can convert this contract to one where instead the value XeR(0,m) is delivered
at time tm by simply depositing the payoff XeR(0,n) in the money market account
up to time tm . The present price is of course the same. Using the theorem in
Lecture Note 1, we have the following equalities from the two contracts:

(tn ) (tm )
P = P (tn ) [XeR(0,n) ] = F0 [X] and P = P (tm ) [XeR(0,m) ] = F0 [X]

By (1) this implies that

b (n) [X] = E
E b (m) [X]

Q.E.D.

Proof of theorem 2
Consider the following strategy: Let A be any event whose outcome is known
at time tj < tk . At time tk−1 enter a long position of eR(0,k) futures contracts if A
has occurred, otherwise do nothing. At tk collect IA (Fk −Fk−1 )eR(0,k) (which may
be negative) and close the contract. This gives the payment IA (Fk − Fk−1 )eR(0,k)
at time tk and no other cash flow. The present price is zero, hence

b A (Fk − Fk−1 )]
0 = P (tk ) [IA (Fk − Fk−1 )eR(0,k) ] = F0 [IA (Fk − Fk−1 )] = E[I

Hence,
b A Fk ] = E[I
E[I b A Fk−1 ]

Employing this equality repeatedly gives, for j < k:

b A Fj ] = E[I
E[I b A Fk ]

Since A can be any event whose outcome is known at time tj , this means that (see
“Conditional Expectations and Martingales” in Lecture Note 7)

b t [Fk ] for j < k.

Fj = E j

This proves theorem 2.

Theorem 3
The following formula for the present prices pj as of time tj for the asset X
to be delivered at time T holds:

b t [pk e−R(j,k) ],
pj = E j < k.
j

In particular,
b t [Xe−R(j,T ) ]
pj = E j

33
Lecture Note 10

Proof
With the notation of Lecture Note 1,
(T ) b t [Xe−R(j,T ) ]
p j = Fj [Xe−R(j,T ) ] = E j

and likewise
b t [Xe−R(k,T ) ]
pk = E k

Now we can employ the iterated expectations rule to get

£ ¤
b t [Xe−R(j,T ) ] = E
pj = E bt E b t [Xe−R(j,T ) ]
j j k
£ ¤
=Eb t e−R(j,k) E
b t [Xe−R(k,T ) ] = Eb t [e−R(j,k) pk ]
j k j

Q.E.D.

34
Lecture Note 11

Note 11: A Model of the Short Interest Rate: Ho-Lee

We divide time into short time intervals t0 , . . . , tn = T , tk −tk−1 = ∆t. Ideally, the
points in time tk should coincide with the times of settlement of futures contracts.
At tk−1 there is an interest rate rk ∆t prevailing from tk−1 to tk . This interest
rate is random, but its value is known at time tk−1 . The Ho-Lee model of the
interest rate is:
√
rk = ak + σ ∆t (z1 + · · · + zk−1 )

where ak are some numbers, the factor σ is the volatility of the short interest rate.
The zj :s are independent N(0,1)-variables and the outcome of each zj occurs at
time tj . This is under the true probability measure.
Under the futures measure, the model looks the same, with σ maintained, but
with new terms ak ; the reasoning is the same as in Lecture Note 8 (“remarks”.)
Thus, under the futures measure
√
rk = θk + σ ∆t (z1 + · · · + zk−1 )

It remains to compute the exact value of θk , but first a notational simplification:

multiplying by ∆t yields
3
rk ∆t = θk ∆t + σ∆t 2 (z1 + · · · + zk−1 )

We now normalise ∆t = 1. This means that rk is the one period interest rate—it
is proportional to ∆t whereas σ is proportional to ∆t3/2 . Now the model reads

rk = θk + σ(z1 + · · · + zk−1 ) (1)

Let Ztk be the price of a zero coupon bond maturing at tk with face value 1. Then,
according to Theorem 3, Lecture Note 10,

b e−r1 −···−rk ]
Ztk = E[1

In order for the model (1) to correctly represent the current term structure, we
must thus have

b −θ1 −···−θk −σ((k−1)z1 +(k−2)z2 +···+1 zk−1 ) ]

Ztk = E[e
b −σ((k−1)z1 +(k−2)z2 +···+1 zk−1 ) ]
= e−θ1 −···−θk E[e

i.e.,
σ2
((k−1)2 +···+12 )
Ztk = e−θ1 −···−θk e 2

Since this must hold for any k, this is equivalent to

³Z ´ σ2
tk−1
θk = ln + (k − 1)2
Zt k 2

35
Lecture Note 11

The Ho-Lee model thus reads

³Z ´ σ2
tk−1
rt = ln + (t − 1)2 + σ(z1 + · · · + zt−1 ) (2)
Zt k 2

under the futures measure.

The Price of a Zero Coupon Bond

Consider a zero coupon bond with face value 1 which matures at time T . We
will now compute its price Z(t, T ) at an earlier time t < T . At this time, the
bond’s value is, by Theorem 3, Lecture Note 10 (the sub index t on E b indicates
conditional expectation as of time t, i.e., the outcomes of zj for j ≤ t are already
known and are regarded as constants):

b t [e−rt+1 −···−rT ]
Z(t, T ) = E
ZT b − σ2 (t2 +···+(T −1)2 )−σ(T −t)(z1 +···+zt )
= Et [e 2
Zt
× e−σ((T −t−1)zt+1 +(T −t−2)zt+2 +···+1 zT −1 ) ]
ZT − σ2 (t2 +···+(T −1)2 ) −σ(T −t)(z1 +···+zt )
= e 2 e
Zt
×E b t [e−σ((T −t−1)zt+1 +(T −t−2)zt+2 +···+1 zT −1 ) ]
ZT −σ(T −t)(z1 +···+zt ) − σ2 (t2 +···+(T −1)2 ) σ2 ((T −t−1)2 +···+12 )
= e e 2 e2
Zt

We compute the sums

σ2 2 σ2
− (t + · · · + (T − 1)2 ) + ((T − t − 1)2 + · · · + 12 )
2 2
T −t−1
σ2 X
=− (t + j)2 − j 2
2 j=0

T −t−1
σ2 X
=− t (2j + t)
2 j=0

σ2
= − (T − 1)(T − t)t
2

Hence,
ZT − σ2 (T −1)(T −t)t −σ(T −t)(z1 +···+zt )
Z(t, T ) = e 2 e (3)
Zt

36
Lecture Note 11

Forward and Futures on a Zero Coupon Bond

We will now compute the futures and forward prices of a contract maturing
at t on a zero coupon bond maturing at T > t with face value 1. The forward
price is easy, it is
ZT
G=
Zt

regardless of the interest rate model. We leave the proof to the reader. The futures
price F0 is

b ZT e− σ2
2
b
F0 = E[Z(t, T )] = E[ (T −1)(T −t)t −σ(T −t)(z1 +···+zt )
e ]
Zt
σ2
= Ge− 2 (T −1)(T −t)t b −σ(T −t)(z1 +···+zt ) ]
E[e
σ2 σ2 σ2
(T −t)2 t
= Ge− 2 (T −1)(T −t)t
e 2 = Ge− 2 (T −t)(t−1)t

Recall that we have normalised ∆t = 1. With no normalisation, we must replace

3
σ by σ∆t 2 and t and T by t∆t−1 and T ∆t−1 respectively, hence

σ2 σ2
(T −t)t2
F0 = Ge− 2 (T −t)(t−∆t)t
−→ Ge− 2 when ∆t → 0.

We see that in contrast to the situation when interest rates are deterministic, the
forward price G and the futures price F0 differ. Since the price of the underlying
asset (the bond) is negatively correlated with the interest rate, the futures price
σ2
(T −t)t2
is lower; in the Ho-Lee model by factor e− 2 .

The Forward Measure

The computation of prices on bond options, for example, is simpler if we use
the forward measure rather than the futures measure. We will see that in the next
section. Let us therefore derive the forward measure for a certain maturity t. We
have:
t
E(t) [X] = Zt−1 P (t) (X) = Zt−1 F0 (Xe−Σ1 rj )
b −Σt1 rj b
= Zt−1 E(Xe ) = E[XW ]
t
where W = Zt−1 e−Σ1 rj

Substituting (2) we get

t
W = eµ−σ Σ1 (t−j)zj

b ] = 1. Note that this is a specification in accordance

for a constant µ such that E[W
with Girsanov’s theorem (Lecture Note 8) with λi = −σ(t−i). We can thus employ
Girsanov’s theorem to get

37
Lecture Note 11

Theorem
The variable zj is N (−σ(t − j), 1) under the forward measure with maturity
at time t ≥ j. Thus we may write zj = wj − σ(t − j) where wj ∈ N (0, 1) under
the forward (t) measure.

In particular, we have from (3) that under the E(t) -measure,

ZT − σ2 (T −1)(T −t)t −σ(T −t)(w1 −σ(t−1)+···+wt−1 −σ·1+wt )

Z(t, T ) = e 2 e
Zt
(4)
ZT − σ2 (T −t)2 t −σ(T −t)(w1 +···+wt )
= e 2 e
Zt

Pricing a European Option on a Zero Coupon Bond

It is now easy to calculate the price of a European option maturing at t on
a zero coupon bond maturing at time T > t. Let F (Z(t, T )) be the value of the
option at maturity. Then its price today is, using (4) and the
√ fact that w1 +· · ·+wt
is a normal distributed variable with standard deviation t,
Z ∞
£ ¡
(t)
¢¤ Zt ¡ ZT − σ2 (T −t)2 t σ(T −t)√t x ¢ − 1 x2
p = Zt E F Z(t, T ) = √ F e 2 e e 2 dx
2π −∞ Zt

But ZZTt = G0 , the forward price of the underlying bond, hence, (also with no
normalisation of time,)
£ ¡ ¢¤
p = Zt E(t) F Z(t, T )
Z ∞
Zt ¡ σ2 2
√ ¢ 1 2
=√ F G0 e− 2 (T −t) t eσ(T −t) t x e− 2 x dx
2π −∞

Note that the same formula may be obtained by Black’s model for bond options
(Lecture Note 6.)

38
Lecture Note 12

Note 12: Ho-Lee’s Binomial Interest Rate Model

When we want to price other than European interest rate derivatives, we need a
numerical procedure, and one such is again a binomial tree. This is constructed
by replacing zk in Lecture Note 11 by binomial variables bj (see Lecture Note 8,
“Binomial Approximation”.) Hence, time is discrete: t0 , t1 , . . . , tn = T , tk −tk−1 =
∆t. The interest rate rk from tk−1 to tk is assumed to be

k
X
rk = θk + σ bj k = 1, . . . , n
2

where {bk } are binomial random variables which takes the values ±1, each with
probability 0.5 under the futures measure; they are thus assumed to be statistically
independent. The outcome of the variable bk occurs at time k − 1. Since we don’t
need the dynamics under the true probability measure, we model the dynamics
from the start under the futures measure. However, as in Lecture Note 11, σ
represents the volatility of the interest rate under either measure, so it can be
3
measured from real data; σ is proportional to ∆t 2 just as in Lecture Note 11.
We will choose the terms θj :s such that the model is consistent with the
current term structure—it will be close to, but not exactly, the same as in the
Normal distribution case.
Let us compute the price Z4 at time t0 of a zero coupon bond maturing at
time t4 with face value 1:

b · e−r1 −r2 −r3 −r4 ] = E[e

Z4 = E[1 b −θ1 −θ2 −θ3 −θ4 −σ(3b2 +2b3 +b4 ) ]
b −3σb2 ] E[e
= e−θ1 −θ2 −θ3 −θ4 E[e b −2σb3 ] E[e
b −σb4 ]
= e−θ1 −θ2 −θ3 −θ4 cosh(3σ) cosh(2σ) cosh(σ)

And, by the same token, in general

¡ ¢
Zk = e−θ1 −···−θk cosh (k − 1)σ · . . . · cosh(σ)

Since this must hold for all k, we must have

³Z ´ £ ¡ ¢¤
k−1
θk = ln + ln cosh (k − 1)σ
Zk

and the dynamics under the futures measure is thus described by

³Z ´ £ ¡ ¢¤ k
X
k−1
rk = ln + ln cosh (k − 1)σ + σ bj k = 1, . . . , n
Zk 2
(
1
bj = 1 with probability 2
where 1
bj = −1 with probability 2

Once we have the parameters of the model, we can price interest rate deriva-
tives in the binomial tree. We show the procedure by an example, where we want

39
Lecture Note 12

to price a European call option maturing at t2 with strike price 86 on a zero

coupon bond maturing at t4 with face value 100 when the following parameters
are given:
¡ f1 = 0.06,
¢ f2 = 0.06095, f3 = 0.06180, f4 = 0.06255, σ = 0.01, where
fk = ln Zk−1 /Zk are the forward rates. We represent the interest rates in a
binomial tree:
t0 t1 t2 t3
0.06 0.071 0.082 0.093
0.051 0.062 0.073
0.042 0.053
0.033
The interest rate from one period to the next is obtained by going either one step
to the right on the same line, or step to the right to the line below; each with a
(risk adjusted) probability of 0.5. We can compute the value at t2 of the bond:
since its value at t4 is 100, the value at t3 and t2 is obtained recursively backwards
by use of the formula of theorem 3 in Lecture Note 10:

t2 t3
84.794 91.119
88.254 92.960
91.856 94.838
96.754

The value of the option can now also be obtained by backward recursion:

t0 t1 t2
2.309 1.050 0
3.853 2.254
5.856

The value of the option is thus 2.309. It is also easy to price other more exotic
derivatives in this binomial tree model.

40