Financial Calculus
An introduction to derivative pricing
Martin Baxter
Nomura International London
Andrew Rennie
Head of Debt Analytics, Merrill Lynch, Europe
Contents
Preface i
The parable of the bookmaker iii
1 Introduction 1
1.1 Expectation pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Arbitrage pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Expectation vs arbitrage . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Discrete processes 7
2.1 The binomial branch model . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 The binomial tree model . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3 Binomial representation theorem . . . . . . . . . . . . . . . . . . . . . 22
2.4 Overture to continuous models . . . . . . . . . . . . . . . . . . . . . . 32
3 Continuous processes 34
3.1 Continuous processes . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.2 Stochastic calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.3 It ˆ o calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.4 Change of measure — the CMG theorem . . . . . . . . . . . . . . . 48
3.5 Martingale representation theorem . . . . . . . . . . . . . . . . . . . . 59
3.6 Construction strategies . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.7 BlackScholes model . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.8 BlackScholes in action . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4 Pricing market securities 77
4.1 Foreign exchange . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.2 Equities and dividends . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.3 Bonds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.4 Market price of risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.5 Quantos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
5 Interest rates 100
5.1 The interest rate market . . . . . . . . . . . . . . . . . . . . . . . . . . 100
5.2 A simple model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
ii
CONTENTS i
5.3 Singlefactor HJM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
5.4 Shortrate models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
5.5 Multifactor HJM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
5.6 Interest rate products . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
5.7 Multifactor models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
6 Bigger models 139
6.1 General stock model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
6.2 Lognormal models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
6.3 Multiple stock models . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
6.4 Numeraires . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
6.5 Foreign currency interestrate models . . . . . . . . . . . . . . . . . . 150
6.6 Arbitragefree complete models . . . . . . . . . . . . . . . . . . . . . 153
A Further reading 157
B Notation 161
C Glossary of technical terms 165
Preface
Notoriously, works of mathematical ﬁnance can be precise, and they can be compre
hensible. Sadly, as Dr Johnson might have put it, the ones which are precise are not
necessarily comprehensible, and those comprehensible are not necessarily precise.
But both are needed. The mathematics of ﬁnance is not easy, and much market
practice is based on a soft understanding of what is actually going on. This is usually
enough for experienced practitioners to price existing contracts, but often insufﬁcient
for innovative new products. Novices, managers and regulators can be left to stumble
around in literature which is ill suited to their need for a clear explanation of the basic
principles. Such ‘seat of the pants’ practices are more suited to the pioneering days of
an industry, rather than the mature $15 trillion market which the derivatives business
has become.
On the academic side, effort is too often expended on ﬁnding precise answers to
the wrong questions. When working in isolation from the market, the temptation
is to ﬁnd analytic answers for their own sake with no reference to the concerns of
practitioners. In particular, the importance of hedging both as a justiﬁcation for the
price and as an important end in itself is often underplayed. Scholars need to be
aware of such ﬁnancial issues, if only because some of the very best work has arisen
in answering the questions of industry rather than academe.
Guide to the chapters
Chapter one is a brief warning, especially to beginners, that the expected worth of
something is not a good guide to its price. That idea has to be shaken off and arbitrage
pricing take its place.
Chapter two develops the idea of hedging and pricing by arbitrage in the discrete
time setting of binary trees. The key probabilistic concepts of conditional expecta
tion, martingales, change of measure, and representation are all introduced in this
simple framework, accompanied by illustrative examples.
Chapter three repeats all the work of its predecessor in the continuous time setting.
Brownian motion is brought out, as well as the It ˆ o calculus needed to manipulate it,
culminating in a derivation of the BlackScholes formula.
Chapter four runs through a variety of actual ﬁnancial instruments, such as div
idend paying equities, currencies and coupon paying bonds, and adapts the Black
Scholes approach to each in turn. A general pattern of the distinction between trad
i
ii PREFACE
able and nontradable quantities leads to the deﬁnition the market price of risk, as
well as a warning not to take that name too seriously. A section on quanto products
provides a showcase of examples.
Chapter ﬁve is about the interest rate market. In spirit, a market of bonds is
much like a market of stocks, but the richness of this market makes it more than
just a special case of BlackScholes. Market models are discussed with a joint short
rate/HJM approach, which lies within the general continuous framework set up in
chapter three. One section details a few of the many possible interest rate contracts,
including swaps, caps/ﬂoors and swaptions. This is a substantial chapter reﬂecting
the depth of ﬁnancial and technical knowledge that has to be introduced in an under
standable way. The aim is to tell one basic story of the market, which all approaches
can slot into.
Chapter six concludes with some technical results about larger and more general
models, including multiple stock nfactor models, stochastic numeraires, and foreign
exchange interestrate models. The running link between the existence of equivalent
martingale measures and the ability to price and hedge is ﬁnally formalized.
A short bibliography, complete answers to the (small) number of exercises, a full
glossary of technical terms and an index are in the appendices.
How to read this book
The book can be read either sequentially as an unfolding story, or by random access
to the selfcontained sections. The occasional questions are to allow practice of the
requisite skills, and are never essential to the development of the material.
A reader is not expected to have any particular prior body of knowledge, except for
some (classical) differential calculus and experience with symbolic notation. Some
basic probability deﬁnitions are contained in the glossary, whereas more advanced
readers will ﬁnd technical asides in the text from time to time.
Acknowledgements
We would like to thank David Tranah at CUP for politely never mentioning the num
ber of deadlines we missed, as well as his much more invaluable positive assistance;
the many readers in London, New York and various universities who have been sub
jected to writing far worse than anything remaining in the ﬁnished edition. Special
thanks to Lorne Whiteway for his help and encouragement.
June 1996
Martin Baxter
Andrew Rennie
The parable of the bookmaker
A
bookmaker is taking bets on a twohorse race. Choosing to be scientiﬁc, he
studies the form of both horses over various distances and goings as well as
considering such factors as training, diet and choice of jockey. Eventually
he correctly calculates that one horse has a 25% chance of winning, and the other a
75% chance. Accordingly the odds are set at 31 against and 31 on respectively.
But there is a degree of popular sentiment reﬂected in the bets made, adding up to
$5 000 for the ﬁrst and $10 000 for the second. Were the second horse to win, the
bookmaker would make a net proﬁt of $1667, but if the ﬁrst wins he suffers a loss
of $5000. The expected value of his proﬁt is 25%×(−$5000) + 75%×($1667) = $0,
or exactly even. In the long term, over a number of similar but independent races,
the law of averages would allow the bookmaker to break even. Until the long term
comes, there is a chance of making a large loss.
Suppose however that he had set odds according to the money wagered — that is,
not 31 but 21 against and 21 on respectively. Whichever horse wins, the bookmaker
exactly breaks even. The outcome is irrelevant.
In practice the bookmaker sells more than 100% of the race and the odds are short
ened to allow for proﬁt (see table). However, the same pattern emerges. Using the
actual probabilities can lead to longterm gain but there is always the chance of a
substantial shortterm loss. For the bookmaker to earn a steady riskless income, he is
best advised to assume the horses’ probabilities are something different. That done,
he is in the surprising position of being disinterested in the outcome of the race, his
income being assured.
A note on odds
When a price is quoted in the form nm against, such as 31 against, it means that
a successful bet of $m will be rewarded with $n plus stake returned. The implied
probability of victory (were the price fair) is m/(m + n). Usually the probability
is less than half a chance so the ﬁrst number is larger than the second. Otherwise,
what one might write as 13 is often called odds of 31 on.
iii
iv THE PARABLE OF THE BOOKMAKER
Actual probability 25% $5000
Bets placed 75% $10 000
1. Quoted odds 135 against 154 on Total = 107%
Implied probability 28% 79% Expected proﬁt = $1 000
Proﬁt if horse wins $3000 $2333
2. Quoted odds 95 against 52 on Total = 107%
Implied probability 36% 71% Expected proﬁt = $1 000
Proﬁt if horse wins $1000 $1000
Allowing the bookmaker to make a proﬁt, the odds change slightly. In the ﬁrst
case, the odds relate to the actual probabilities of a horse winning the race. In the
second, the odds are derived from the amounts of money wagered.
Chapter 1
Introduction
F
inancial market instruments can be divided into two distinct species. There
are the ‘underlying’ stocks: shares, bonds, commodities, foreign currencies;
and their ‘derivatives’, claims that promise some payment or delivery in the
future contingent on an underlying stock’s behavior. Derivatives can reduce risk —
by enabling a player to ﬁx a price for a future transaction now, for example — or
they can magnify it. A costless contract agreeing to pay off the difference between
a stock and some agreed future price lets both sides ride the risk inherent in owning
stock without needing the capital to buy it outright.
In form, one species depends on the other — without the underlying (stock) there
could be no future claims — but the connection between the two is sufﬁciently com
plex and uncertain for both to trade ﬁercely in the same market. The apparently
random nature of stocks ﬁlters through to the claims — they appear random too.
Yet mathematicians have known for a while that to be random is not necessarily
to be without some internal structure — put crudely, things are often random in non
random ways. The study of probability and expectation shows one way of coping
with randomness and this book will build on probabilistic foundations to ﬁnd the
strongest possible links between claims and their random underlying stocks. The
current state of truth is, however, unfortunately complex and there are many false
trails through this zoo of the new. Of these, one is particularly tempting.
1.1 Expectation pricing
Consider playing the following game — someone tosses a coin and pays you one
dollar for heads and nothing for tails. What price should you pay for this prize? If
the coin is fair, then heads and tails are equally likely — about half the time you
should win the dollar and the rest of the time you should receive nothing. Over
enough plays, then, you expect to make about ﬁfty cents a go. So paying more than
ﬁfty cents seems extravagant and less than ﬁfty cents looks extravagant for the person
offering the game. Fifty cents, then, seems about right.
Fifty cents is also the expected proﬁt from the game under a more formal, mathe
matical deﬁnition of expectation. A probabilistic analysis of the game would observe
1
2 CHAPTER 1. INTRODUCTION
that although the outcome of each coin toss is essentially random, this is not inconsis
tent with a deeper nonrandom structure to the game. We could posit that there was
a ﬁxed measure of likelihood attached to the coin tossing, a probability of the coin
landing heads or tails of
1
2
. And along with a probability ascription comes the idea
of expectation, in this discrete case, the total of each outcome’s value weighted by its
attached probability. The expected payoff in the game is
1
2
×$1 +
1
2
×$0 = $0.50.
This formal expectation can then be linked to a ‘price’ for the game via something
like the following:
Kolmogorov’s strong law of large numbers
Suppose we have a sequence of independent random numbers X
1
, X
2
, X
3
, and so
on, all sampled from the same distribution, which has mean (expectation) µ, and
we let S
n
be the arithmetical average of the sequence up to the nth term, that is
S
n
= (X
1
+X
2
+. . . +X
n
)/n. Then, with probability one, as n gets larger the value
of S
n
tends towards the mean µ of the distribution.
If the arithmetical average of outcomes tends towards the mathematical expecta
tion with certainty, then the average proﬁt/loss per game tends towards the mathe
matical expectation less the price paid to play the game. If this difference is positive,
then in the long run it is certain that you will end up in proﬁt. And if it is negative,
then you will approach an overall loss with certainty. In the short term of course,
nothing can be guaranteed, but over time, expectation will out. Fifty cents is a fair
price in this sense.
But is it an enforceable price? Suppose someone offered you a play of the game
for 40 cents in the dollar, but instead of allowing you a number of plays, gave you
just one for an arbitrarily large payoff. The strong law lets you take advantage of
them over repeated plays: 40 cents a dollar would then be ﬁnancial suicide, but it
does nothing if you are allowed just one play. Mortgaging your house, selling off all
your belongings and taking out loans to the limit of your credit rating would not be a
rational way to take advantage of this source of free money.
So the ‘market’ in this game could trade away from an expectation justiﬁed price.
Any price might actually be charged for the game in the short term, and the number
of ‘buyers’ or ‘sellers’ happy with that price might have nothing to do with the math
ematical expectation of the game’s outcome. But as a guide to a starting price for the
game, a ballpark amount to charge, the strong law coupled with expectation seems
to have something going for it.
Time value of money
We have ignored one important detail — the time value of money. Our analysis of the
coin game was simpliﬁed by the payment for and the payoff from the game occurring
at the same time. Suppose instead that the coin game took place at the end of a year,
but payment to play had to be made at the beginning — in effect we had to ﬁnd the
1.1. EXPECTATION PRICING 3
value of the coin game’s contingent payoff not as of the future date of play, but as of
now.
If we are in January, then one dollar in December is not worth one dollar now, but
something less. Interest rates are the formal acknowledgement of this, and bonds are
the market derived from this. We could assume the existence of a market for these
future promises, the prices quoted for these bonds being structured, derivable from
some interest rate. Speciﬁcally:
Time value of money
We assume that for any time T less than some time horizon r, the value now of a
dollar promised at time T is given by exp(−rT) for some constant r > 0. The rate
r is then the continuously compounded interest rate for this period.
The interest rate market doesn’t have to be this simple; r doesn’t have to be con
stant. And indeed in real markets it isn’t. But here we assume it is. We can derive a
stronglaw price for the game played at time T. Paying 50 cents at time T is the same
as paying 50 exp(−rT) cents now. Why? Because the payment of 50 cents at time
T can be guaranteed by buying half a unit of the appropriate bond (that is, promise)
now, for cost 50 exp(−rT) cents. Thus the stronglaw price must be not 50 cents but
50 exp(−rT) cents.
Stocks, not coins
What about real stock prices in a real ﬁnancial market? One widely accepted model
holds that stock prices are lognormally distributed. As with the time value of money
above, we should formalize this belief.
Stock model
We assume the existence of a random variable X, which is normally distributed
with mean µ and standard deviation σ, such that the change in the logarithm of the
stock price over some time period T is given by X. That is
log S
T
= log S
0
+ X or S
T
= S
0
exp(X).
Suppose, now, that we have some claim on this stock, some contract that agrees
to pay certain amounts of money in certain situations — just as the coin game did.
The oldest and possibly most natural claim on a stock is the forward: two parties
enter into a contract whereby one agrees to give the other the stock at some agreed
point in the future in exchange for an amount agreed now. The stock is being sold
forward. The ‘pricing question’ for the forward stock ‘game’ is: what amount should
be written into the contract now to pay for the stock one year in the future?
We can dress this up in formal notation — the stock price at time T is given
by S
T
, and the forward payment written into the contract is K, thus the value of
the contract at its expiry, that is when the stock transfer actually takes place, is
4 CHAPTER 1. INTRODUCTION
S
T
− K. The time value of money tells us that the value of this claim as of now
is exp(−rT)(S
T
− K). The strong law suggests that the expected value of this ran
dom amount, E(exp(−rT)(S
T
− K)), should be zero. If it is positive or negative,
then longterm use of that pricing should lead to one side’s proﬁt. Thus one ap
parently reasonable answer to the pricing question says K should be set so that
E(exp(−rT)(S
T
−K)) = 0, which happens when K = E(S
T
).
What is E(S
T
)? We have assumed that log(S
T
) − log(S
0
) is normally distributed
with mean µ and variance σ
2
— thus we want to ﬁnd E(S
0
exp(X)), where X is nor
mally distributed with mean µ and standard deviation σ. For that, we can use a result
such as:
The law of the unconscious statistician
Given a realvalued randomvariable X with probability density function f(x), then
for any integrable real function h, the expectation of h(X) is
E(h(X)) =
_
∞
−∞
h(x)f(x)dx.
Since X is normally distributed, the probability density function for X is
f(x) =
1
√
2πσ
2
exp
_
−(x −µ)
2
2σ
2
_
.
Integration and the law of the unconscious statistician then tells us that the expected
stock price at time T is S
0
exp
_
µ +
1
2
σ
2
_
. This is the stronglaw justiﬁed price for
the forward contract; just as with the coin game, it can only be a suggestion as to
the market’s trading level. But the technique will clearly work for more than just
forwards. Many claims are capable of translation into functional form, h(X), and the
law of the unconscious statistician should be able to deliver an expected value for
them. Discounting this expectation then gives a theoretical value which the strong
law tempts us into linking with economic reality.
1.2 Arbitrage pricing
So far, so plausible — but seductive though the strong law is, it is also completely
useless. The price we have just determined for the forward could only be the market
price by an unfortunate coincidence. With markets where the stock can be bought and
sold freely and arbitrary positive and negative amounts of stock can be maintained
without cost, trying to trade forward using the strong law would lead to disaster — in
most cases there would be unlimited interest in selling forward to you at that price.
Why does the strong law fail so badly with forwards? As mentioned above in the
context of the coin game, the strong law cannot enforce a price, it only suggests. And
in this case, another completely different mechanism does enforce a price. The fair
price of the contract is S
0
exp(rT). It doesn’t depend on the expected value of the
1.3. EXPECTATION VS ARBITRAGE 5
stock, it doesn’t even depend on the stock price having some particular distribution.
Either counterparty to the contract can in fact construct the claim at the start of the
contract period and then just wait patiently for expiry to exchange as appropriate.
Construction strategy
Consider the seller of the contract, obliged to deliver the stock at time T in exchange
for some agreed amount. They could borrow S
0
now, buy the stock with it, put
the stock in a drawer and just wait. When the contract expires, they have to pay
back the loan — which if the continuously compounded rate is r means paying back
S
0
exp(rT), but they have the stock ready to deliver. If they wrote less than S
0
exp(rT)
into the contract as the amount for forward payment, then they would lose money with
certainty.
So the forward price is bounded below by S
0
exp(rT). But of course, the buyer
of the contract can run the scheme in reverse, thus writing more than S
0
exp(rT)
into the contract would guarantee them a loss. The forward price is bounded above
by S
0
exp(rT) as well. Thus there is an enforced price, not of S
0
exp
_
µ +
1
2
σ
2
_
but
S
0
exp(rT). Any attempt to strike a different price and offer it into a market would
inevitably lead to someone taking advantage of the free money available via the con
struction procedure. And unlike the coin game, mortgaging the house would now be
a rational action. This type of market opportunism is old enough to be ennobled with
a name — arbitrage. The price of S
0
exp(rT) is an arbitrage price — it is justiﬁed be
cause any other price could lead to unlimited riskless proﬁts for one party. The strong
law wasn’t wrong — if S
0
exp
_
µ +
1
2
σ
2
_
is greater than S
0
exp(rT), then a buyer of a
forward contract expects to make money. (But then of course, if the stock is expected
to grow faster than the riskless interest rate r, so would buyers of the stock itself.)
But the existence of an arbitrage price, however surprising, overrides the strong law.
To put it simply, if there is an arbitrage price, any other price is too dangerous to
quote.
1.3 Expectation vs arbitrage
The strong law and expectation give the wrong price for forwards. But in a certain
sense, the forward is a special case. The construction strategy — buying the stock
and holding it — certainly wouldn’t work for more complex claims. The standard
call option which offers the buyer the right but not the obligation to receive the stock
for some strike price agreed in advance certainly couldn’t be constructed this way. If
the stock price ends up above the strike, then the buyer would exercise the option and
ask to receive the stock — having it salted away in a drawer would then be useful to
the seller. But if the stock price ends up below the strike, the buyer will abandon the
option and any stock owned by the seller would have incurred a pointless loss.
Thus maybe a stronglaw price would be appropriate for a call option, and until
1973, many people would have agreed. Almost everything appeared safe to price via
6 CHAPTER 1. INTRODUCTION
expectation and the strong law, and only forwards and close relations seemed to have
an arbitrage price. Since 1973, however, and the infamous BlackScholes paper, just
how wrong this is has slowly come out. Nowhere in this book will we use the strong
law again. Just to muddy the waters, though, expectation will be used repeatedly, but
it will be as a tool for riskfree construction. All derivatives can be built from the
underlying — arbitrage lurks everywhere.
Chapter 2
Discrete processes
T
he goal of this book is to explore the limits of arbitrage. Bit by bit we will
put together a mathematical framework strong enough to be a realistic model
of the real ﬁnancial markets and yet still structured enough to support con
struction techniques. We have a long way to go, though; it seems wise to start very
small.
2.1 The binomial branch model
Something random for the stock and something to represent the timevalue of money.
At the very least we need these two things — any model without them cannot begin to
claim any relation to the real ﬁnancial market. Consider, then, the simplest possible
model with a stock and a bond.
The stock
Just one timetick — we start at time t = 0 and end a short tick later at time t = δt. We
need something to represent the stock, and it had better have some unpredictability,
some random component. So we suppose that only two things can happen to the
stock in this time: an ‘up’ move or a ‘down’ move. With just two things allowed to
happen, pictorially we have a branch (ﬁgure 2.1).
Figure 2.1: The binomial branch
Our randomness will have some structure — we will assign probabilities to the up
7
8 CHAPTER 2. DISCRETE PROCESSES
and down move: probability p to move up to node 3, and thus 1 −p to move down to
node 2. The stock will have some value at the start (node 1 as labeled on the picture),
call it s
1
. This value represents a price at which we can buy and sell the stock in
unlimited amounts. We can then hold on to the stock across the time period until
time t = δt. Nothing happens to us in the intervening period by dint of holding on to
the stock — there is no charge for holding positive or negative amounts — but at the
end of the period it will have a new value. If it moves down, to node 2, then it will
have value s
2
; up, to node 3, value s
3
.
The bond
We also need something to represent the timevalue of money — a cash bond. There
will be some continuously compounded interest rate r that will hold for the period
t = 0 to t = δt — one dollar at time zero will grow to $ exp(rδt). We should be able to
lend at that rate, and borrow — and in arbitrary size. To represent this, we introduce
a cash bond B which we can buy or sell at time zero for some price, say B
0
, and
which will be worth a deﬁnite B
0
exp(rδt) a tick later.
These two instruments are our ﬁnancial world, and simple though it is it still has
uncertainties for investors. Only one of the possible stock values might suit a partic
ular player, their plans surviving or failing by the random outcome. Thus there could
be a market for instruments dependent on the value the stock takes at the end of the
tickperiod. The investor’s requirement for compensation based on the future value
of the stock could be codiﬁed by a function f mapping the two future possibilities,
node 2 or node 3, to two rewards or penalties f(2) and f(3). A forward contract,
struck at k, for example, could be codiﬁed as f(2) = s
2
−k, f(3) = s
3
−k.
Riskfree construction
The question can now be posed — exactly what class of functions f can be explicitly
constructed via a suitable strategy? Clearly the forward can be — as in chapter one,
we would buy the stock (cost: s
1
), and sell off cash bonds to fund the purchase. At the
end of the period, we would be able to hand over the stock and demand s
1
exp(rδt) in
exchange. The price k of the forward thus has to be s
1
exp(rδt) exactly as we would
have hoped — priced via arbitrage.
But what about more complex f? Can we still ﬁnd a construction strategy? Our
ﬁrst guess would be no. The stock takes one of two random values at the end of the
tickperiod and the value of the derivative would in general be different as well. The
probabilities of each outcome for the derivative f are known, thus we also know the
expected value of f at the end of the period as well: (1 −p)f(2) +pf(3), but we don’t
know its actual value in advance.
2.1. THE BINOMIAL BRANCH MODEL 9
Bondonly strategy
All is not lost, though. Consider a portfolio of just the cash bond. The cash bond
will grow by a factor of exp(rδt) across the period, thus buying discount bonds to the
value of exp(−rδt)[(1 −p)f(2) + pf(3)] at the start of the period will provide a value
equal to (1 −p)f(2) +pf(3) at the end. Why would we choose this value as the target
to aim for? Because it is the expected value of the derivative at the end of the period
— formally:
Expectation for a branch
Let S be a binomial branch process with base value s
1
at time zero, downvalue s
2
and upvalues s
3
. Then the expectation of S at ticktime 1 under the probability of
an upmove p is:
E
p
(S
1
) = (1 −p)s
2
+ ps
3
Our claimf on S is just as much a randomvariable as S
1
is —we can meaningfully
talk of its expectation. And thus we can meaningfully aim for the expectation of the
claim, via the cash bonds. This strategy of construction would at the very least be
expected to break even. And the value of the starting portfolio of cash bonds might be
claimed to be a good predictor of the value of the derivative at the start of the period.
The price we would predict for the derivative would be the discounted expectation of
its value at the end.
But of course this is just the strong law of chapter one all over again — just thinly
disguised as construction. And exactly as before we are missing an element of co
ercion. We haven’t explicitly constructed the two possible values the derivative can
take: f(2) and f(3); we have simply aimed between them in a probabilistic sense and
hoped for the best.
And we already know that this best isn’t good enough for forwards. For a stock
that obeys a binomial branch process, its forward price is not suggested by the pos
sible stock values s
2
and s
3
, but enforced by the interest rate r implied by the cash
bond B: namely s
1
exp(rδt). The discounted expectation of the claim doesn’t work
as a pricing tool.
Stocks and bonds together
But can we do any better? Another strategy might occur to us, we have after all two
instruments which we can build into a portfolio to hold for the tickperiod. We tried
using the guaranteed growth of the cash bond as a device for producing a particular
desired value, and we chose the expected value of the derivative as our target point.
But we have another instrument tied more strongly to the behavior of both the stock
and the derivative than just the cash bond. Namely the stock itself, Suppose we
attempted to guarantee not an amount known in advance which we hope will stand as
a reasonable predictor for the value of the derivative, but the value of the derivative
10 CHAPTER 2. DISCRETE PROCESSES
itself, whatever it might be.
Consider a general portfolio (φ, ψ), namely φ of the stock S (worth φs
1
) and ψ of
the cash bond B (worth ψB
0
). If we were to buy this portfolio at time zero, it would
cost φs
1
+ ψB
0
.
One tick later, though, it would be worth one of two possible values:
φs
3
+ ψB
0
exp(rδt) after an ‘up’ move,
and φs
2
+ ψB
0
exp(rδt) after an ‘down’ move.
This pair of equations should intrigue us — we have two equations, two possible
claim values and two free variables φ and ψ. We have two values f(3) and f(2)
which we want to duplicate under the appropriate move of the stock, and we have
two variables φ and ψ which we can adjust. Thus the strategy can reduce to solving
the following two simultaneous equations for (φ, ψ):
φs
3
+ ψB
0
exp(rδt) = f(3),
φs
2
+ ψB
0
exp(rδt) = f(2).
Except if perversely s
2
and s
3
are identical — in which case S is a bond not a stock
— we have the solutions:
φ =
f(3) −f(2)
s
3
−s
2
,
ψ = B
−1
0
exp(−rδt)
_
f(3) −
(f(3) −f(2))s
3
s
3
−s
2
_
.
What can we do with this algebraic result? If we bought this (φ, ψ) portfolio and
held it, the equations guarantee that we achieve our goal — if the stock moves up,
then the portfolio becomes worth f(3); and if the stock moves down, the portfolio
becomes worth f(2). We have synthesized the derivative.
The price is right
Our simple model allows a surprisingly prescient strategy. Any derivative f can be
constructed from an appropriate portfolio of bond and stock. And constructed in
advance. This must have some effect on the value of the claim, and of course it
does — unlike the expectation derived value, this is enforceable in an ideal market
as a rational price. Denote by V the value of buying the (φ, ψ) portfolio, namely
φs
1
+ ψB
0
, which is:
V = s
1
_
f(3) −f(2)
s
3
−s
2
_
+ exp(−rδt)
_
f(3) −
(f(3) −f(2))s
3
s
3
−s
2
_
Now consider some other market maker offering to buy or sell the derivative for a
price P less than V . Anyone could buy the derivative from them in arbitrary quantity,
and sell the (φ, ψ) portfolio to exactly match it. At the end of the tickperiod the value
of the derivative would exactly cancel the value of the portfolio, whatever the stock
2.1. THE BINOMIAL BRANCH MODEL 11
price was — thus this set of trades carries no risk. But the trades were carried out at a
proﬁt of V −P per unit of derivative/portfolio bought — by buying arbitrary amounts,
anyone could make arbitrary riskfree proﬁts. So P would not have been a rational
price for the market maker to quote and the market would quickly have mobilized to
take advantage of the ‘free’ money on offer in arbitrary quantity.
Similarly if a market maker quoted the derivative at a price P greater than V ,
anyone could sell them it and buy the (φ, ψ) portfolio to lock in a riskfree proﬁt of
P −V per unit trade. Again the market would take advantage of the opportunity.
Only by quoting a twoway price of V can the market maker avoid handing out
riskfree proﬁts to other players — hence V is the only rational price for the derivative
at time zero, the start of the tickperiod. Our model, though allowing randomness,
lets arbitrage creep everywhere — the strong law can be banished completely.
Example — the whole story in one step
We have an interestfree bond and a stock, both initially priced at $1. At the end of
the next time interval, the stock is worth either $2 or $0.50. What is the worth of a
bet which pays $1 if the stock goes up?
Solution. Let B denote the bond price, S the stock price, and X the payoff of the bet.
The picture describes the situation:
Figure 2.2: Pricing a bet
Buy a portfolio consisting of 2/3 of a unit of stock and a borrowing of 1/3 of a unit
of bond, The cost of this portfolio at time zero is
2
3
× $1 −
1
3
× $1 = $0.33. But after
an upjump, this portfolio becomes worth
2
3
×$2 −
1
3
×$1 = $1. After a downjump,
it is worth
2
3
× $0.5 −
1
3
× $1 = $0. The portfolio exactly simulates the bet’s payoff,
so has to be worth exactly the same as the bet. It must be that the portfolio’s initial
value of $0.33 is also the bet’s initial value.
Expectation regained
A surprise still lurks. The stronglaw approach may be useless in this model —
leaving aside coincidence, expectation pricing involving the probabilities p and 1 −p
leads to riskfree proﬁts being available. But with an eye to rearranging the equations,
12 CHAPTER 2. DISCRETE PROCESSES
we can deﬁne a simplifying variable:
q =
s
1
exp(rδt) −s
2
s
3
−s
2
.
What can we say about q? Without loss of generality, we can assume that s
3
is
bigger than s
2
. Were q to be less than or equal to 0, then s
1
exp(rδt) ≤ s
2
< s
3
.
But s
1
exp(rδt) is the value that would be obtained by buying s
1
worth of the cash
bond B at the start of the tickperiod. Thus the stock could be bought in arbitrary
quantity ﬁnanced by selling the appropriate amount of cash bond and a guaranteed
riskfree proﬁt made. It is not unreasonable then to eliminate this possibility by ﬁat
— specifying the structure of our market to avoid it. So for any market in which we
have a stock which obeys a binomial branch process S, we have q > 0.
Similarly were q to be greater than or equal to 1, then s
2
< s
3
≤ s
1
exp(rδt) —
and this time selling stock and buying cash bonds provides unlimited riskfree gains.
Thus the structure of a rational market will force q into (0, 1), the interval of points
strictly between 0 and 1 — the same constraint we might demand for a probability.
Now the surprise: when we rewrite the formula for the value V of the (φ, ψ) port
folio (try it) we get:
V = exp(−rδt)
_
(1 −q)f(2) +qf(3)
_
.
Outrageous though it might seem, this is the expectation of the claim under q. This
reappearance of the expectation operator is unsettling.
The price V is not the expected future value of the derivative (discounted suitably
by the growth of the cash bond) — that would involve p in the above formula. Yet V
is the discounted expectation with respect to some number q in (0, 1). If we view the
expectation operator as implying some information about the future — a stronglaw
average over many trials, for example — then V is not what we would unconsciously
call the expected value. It sounds pedantic to say it, but V is an expectation, not an
expected value. And it is easy enough to check that this expectation gives the correct
strike for a forward contract: s
1
exp(rδt).
xxxx; TNxxx
Exercise 2.1 Show that a forward contract, struck at k, can be thought of as
the payoff f, where f(2) = s
2
− k and f(3) = s
3
− k. Now verify, using the
formula for V , that the correct strike price is indeed s
1
exp(rδt).
2.2 The binomial tree model
From branch to tree. Our single time step was simple to analyze, but it represents a
bare minimum as a model. It had a random stock and a cash bond, but it only allowed
the stock two possible values at the end of a single time period. Markets are not quite
that straightforward. But if we could build the branch model up into something more
sophisticated, then we could transfer its results into a larger, better model. This is
2.2. THE BINOMIAL TREE MODEL 13
the intention of this section — we shall build a tree out of branches, and see what
survives.
Our ﬁnancial world will again be just two instruments — a discount bond B and
a stock S. Unlimited amounts of either can be bought and sold without transaction
costs, default risks, or bidoffer spreads. But now, instead of a single timeperiod, we
will allow many, stringing the individual δts together.
The stock
Changes in the value of the stock S must be random — the market demands that —
but the randomness can have structure. Our ministock from the binomial branch
model allowed the stock to change to just two values at the end of the time period,
and we shall keep that structure. But now, we will string these choices together into
a tree. The very ﬁrst time period, from t = 0 to t = δt, will be just as before (a tree of
branches starts with just one simple branch). If the value of S at time zero is S
0
= s
1
,
then the actual value one tick later is not known but the range of possibilities is −S
1
has only two possible values: s
2
and s
3
.
Now, we must extend the branch idea in a natural fashion. One tick δt later still,
the stock again has two possibilities, but dependent on the value at ticktime 1; hence
there are four possibilities. From s
2
, S
2
can be either s
4
or s
5
; from s
3
, S
2
can be
either s
6
or s
7
.
As the picture suggests, at ticktime i, the stock can have one of 2
i
possible values,
though of course given the value at ticktime (i−1), there are still only two admissible
possibilities: from node j the process either goes down to node 2j or up to node 2j+1.
Figure 2.3
This tree arrangement gives us considerably more ﬂexibility. A claim can now
call on not just two possibilities, but any number. If we think that a thousand random
possible values for a stock is a suitable level of complexity, then we merely have to
set δt small enough that we get ten or so layers of the tree in before the claim time
t. We also have a richer allowed structure of probability. Each up/down choice will
14 CHAPTER 2. DISCRETE PROCESSES
have an attached probability of it being made. From the standpoint of notation, we
can represent this pair of probabilities (which must sum to 1) by just one of them
(the up probability) p
j
, the probability of the stock achieving value s
2j+1
, given its
previous value of s
j
. The probability of the stock moving down, and achieving value
s
2j
, is then 1 −p
j
. Again this is shown in the picture.
The cash bond
To go with our grownup stock, we need a grownup cash bond. In the simple branch
model, the cash bond behaved entirely predictably; there was a known interest rate r
which applied across the period making the cash bond price increase by a factor of
exp(rδt). There is no reason to impose such a strict condition — we don’t have to
have a constant interest rate known for the entire tree in advance but instead we could
have a sequence of interest rates, R
0
, R
1
, . . ., each known at the start of the appropri
ate tick period. The value of the cash bond at time nδt thus be B
0
exp
_
n−1
i=0
R
i
δt
_
.
It is worth contrasting the cash bond and the stock. We have admitted the pos
sibility of randomness in the cash bond’s behavior (though in fact we will not yet
be particularly interested in its exact form). But compared to the stock it is a very
different sort of randomness. The cash bond B has the same structure as the time
value of money. The interest which must be paid or earned on cash can change over
time, but the value of a cash holding at the next tick point is always known, because
it depends only on the interest rate already known at the start of the period.
But for simplicity’s sake, we will now keep a constant interest rate r applying
everywhere in the tree, and in this case the price of the cash bond at time nδt is
B
0
exp(rnδt).
Trees are complex
At this stage, the binomial structure of the tree may seem rather arbitrary, or indeed
unnecessarily simplistic. A tree is better than a single branch, but it still won’t allow
continuous ﬂuid changes in stock and bond values. In fact, as we shall see, it more
than suits our purpose. Our ﬁnal goal, an understanding of the limitations (or lack
of them) of riskfree construction when the underlying stocks take continuous values
in continuous time, will draw directly and naturally on this starting point. And as δt
tends to zero, this model will in fact be more than capable of matching the models
we have in mind. Perhaps more pertinently, before we abandon the tree as simplistic,
we had better check that it hasn’t become too complex for us to make any analytic
progress at all.
Backwards induction
In fact most of the hard work has already been done when we examined the branch
model. Extending the results and intuitions of section 2.1 to an entire binomial tree is
2.2. THE BINOMIAL TREE MODEL 15
surprisingly straightforward. The key idea is that of backwards induction — extend
ing the construction portfolio back one tick at a time from the claim to the required
starting place.
Consider, then, a general claim for our stock S. When we examined a single
branching of our tree, we had the function f dependent only on the node chosen at
the end of a single tick period — here we can extend the idea of a claim to cover not
only the value of S at the time the claim is exercised but also the history of S up until
that point.
The tree structure of the stock was not entirely arbitrary — it embodies a one
toone relationship between a node and the history of the stock’s path up to and
including that node. No other history reaches that node; and trivially no other node
is reached by that history. This is precisely that condition that allows us actually to
associate a claim value with a particular endnode on our tree. We shall also insist on
the ﬁniteness of our tree. There must be some ﬁnal ticktime at which the claim is
fully determined. A condition not unreasonable in the real ﬁnancial world. A general
claim can be thought of as some function on the nodes at this claim timehorizon.
The twostep
We know that the expectation operator can be made to work for a single branch —
here, then, we must wade through the algebra for two timesteps, three branches stuck
together into a tree. If two timesteps work, then so will many.
Figure 2.4: Double fork at time 0
Suppose that the interest rate over any branch is constant at rate r. Then there
exists some set of suitable q
j
s such that the value of the derivative at node j at tick
time i, f(j), is
f(j) = e
−rδt
_
q
j
f(2j + 1) + (1 −q
j
)f(2j)
_
.
That is the discounted expectation under q
j
of the time — (i+1) claim values f(2j+1)
and f(2j). So in our twostep tree (ﬁgure 2.4), the two forks from node 3 to nodes 6
and 7, and from node 2 to nodes 4 and 5, are both structurally identical to the simple
16 CHAPTER 2. DISCRETE PROCESSES
onestep branch. This means that f(3) comes from f(6) and f(7) via
f(3) = e
−rδt
_
q
3
f(7) + (1 −q
3
)f(6)
_
,
and similarly, f(2) comes from f(4) and f(5), with
f(2) = e
−rδt
_
q
2
f(5) + (1 −q
2
)f(4)
_
.
Here q
j
is the probability
_
s
j
exp(rδt) −s
2j
_
/(s
2j+1
−s
2j
), so for instance
q
2
=
s
2
exp(rδt) −s
4
s
5
−s
4
, and q
3
=
s
3
exp(rδt) −s
6
s
7
−s
6
.
But now we have a value for the claim at time 1; it is worth f(3) if the ﬁrst jump was
up, and f(2) if it was down. But this initial fork from node 1 to nodes 2 and 3 also
has the single branch structure. Its value at time zero must be
f(1) = e
−rδt
_
q
1
f(3) + (1 −q
1
)f(2)
_
.
Thus the value of the claim at time zero has the daunting looking expression formed
by combining the three equations above,
f(1) = e
−2rδt
_
q
1
q
3
f(7) +q
1
(1 −q
3
)f(6) + (1 −q
1
)q
2
f(5) + (1 −q
1
)(1 −q
2
)f(4)
_
.
We haven’t formally deﬁned expectation on our tree, but it is clear what it must be.
Path probabilities
The probability that the process follows a particular path through the tree is just
the product of the probabilities of each branch taken. For example, in ﬁgure 2.4,
the chance of going up twice is the product q
1
q
3
, the chance of going up and then
down is q
1
(1 −q
3
), and so on.
This is a case of the more general slogan that when working with indepen
dent events, the probabilities multiply.
Expectation on a tree
The expectation of some claim on the ﬁnal nodes of a tree is the sum over those
nodes of the claim value weighted by the probabilities of paths reaching it.
A twostep tree has four possible paths to the end. But each path carries two
probabilities attached to it, one for the ﬁrst time step and one for the second, thus the
pathprobability, the probability of following any particular path, must be the product
of these.
The expectation of a claim is then the total of the four outcomes each weighted
by this pathprobability. But examine the expression we have derived above — it
is of course precisely the expectation of the claims f(7), . . . , f(4), discounted by the
appropriate interestrate factor e
−2rδt
, under the probabilities q
1
q
3
, q
1
(1 − q
3
), (1 −
q
1
)q
2
, (1 −q
1
)(1 −q
2
) corresponding to the ‘probability tree’ (q
1
, q
2
, q
3
).
For claim pricing and expectation, a twostep tree is simply three branches. And
so on.
2.2. THE BINOMIAL TREE MODEL 17
The inductive step
Returning to our general tree over n periods, we start at its ﬁnal layer. All nodes here
have claim values and are in pairs, the ends of single branchings. Consider any one
of these ﬁnal branchings, from a node at time (n − 1) to two nodes at time n. The
results from section 2.1 provide a riskfree construction portfolio (φ, ψ) of stock and
bond at the root of the branch that can generate the time n claim amount. (Both our
grownup stock and the cash bond are indistinguishable over a single branching from
the stock and bond of the simple model.)
Thus the nodes at time (n−1) are all roots of branches that end on the claim layer
and have arbitrage guaranteed values for the derivative attached — claimvalues in
their own right now insisted on not by the investor’s contract (that only applies to the
ﬁnal layer) but by arbitrage considerations. Thus we can work back from enforced
claims at the ﬁnal layer to equally strongly enforced claimvalues at the layer before.
This is the inductive step — we have moved the claims on the ﬁnal layer back one
step.
The inductive result
By repeating the inductive step, we will sweep backwards through the tree. Each
layer will ﬁx the value of the derivative on the layer before, because each layer is
only separated from the layer before by simple branches. What we have done is
essentially a recursive ﬁlling in process. The investor ﬁlled in the nodes at the end of
the tree with claims — we ﬁlled in the rest by constructing (φ, ψ) portfolios at each
branching which guaranteed the correct outcome at the next step.
We will reach the root of the entire tree with a single value. This is the timezero
value of the ﬁnal derivative claim — why? Because just as for the single branch,
there is a construction portfolio which, though it will change at each tick time, will
inexorably lead us to the claim payoff required, whatever path the stock actually
takes.
We now have some idea of the complexity of the construction portfolios that will
be required. Instead of a single amount of stock φ, we now have a whole number
of them, one per node. And as fate casts the die and the stock jumps on the tree,
so this amount will jump as well. Perverse though it may seem for a guaranteed
construction procedure, the construction portfolios (φ
i
, ψ
i
) are also random, just like
the stock. But there is a vital structural difference — they are known just in time to
be useful, unlike the stock value they are known onestep in advance.
Arbitrage has worked its way into the tree model as well. The fact that the tree
is simply lots of branches was enough to banish the strong law here as well. All
claims can be constructed from a stock and bond portfolio, and thus all claims have
an arbitrage price.
18 CHAPTER 2. DISCRETE PROCESSES
Expectation again
The strong law may be useless, but what about expectation? We had no need of the
probabilities p
j
, but the reemergence of the expectation operator is not just a coin
cidence peculiar to the simplicities of the branch model. Yet again the expectation
operator will appear with the correct result — just as the conclusion from the previ
ous section was that with respect to a suitable ‘probability’, the expectation operator
provided the correct local hedge, here we will see that the expectation operator with
respect to some suitable set of ‘probabilities’ also provides the correct global struc
ture for a hedge.
A worked example
We can give a concrete demonstration of how this works. The tree in Figure 2.5 is
called recombinant as different branches can come back together, or recombine, at
the same node. Such trees are computationally much easier to work with, as long as
we remember that there is more than one path to the ﬁnal nodes. The tree nodes are
the stock prices, s, and at each node the process will go up with probability 3/4 and
down with probability 1/4. (For simplicity, interest rates are zero.)
Figure 2.5: A stock price on a recombinant tree
What is the value of an option to buy the stock for 100 at time 3?
It is easy to ﬁll in the value of the claim on the time 3 column. Reading from top
to bottom, the claim has values then of 60, 20, 0 and 0.
We shall now need our equations for the new probabilities q and the claim values
f. As the interest rate r is zero, these equations are a little simpler. If we are about to
move either ‘up’ or ‘down’, then the (riskneutral) probability q is
q =
s
now
−s
down
s
up
−s
down
and the value of a claim, f, now is
f
now
= qf
up
+ (1 −q)f
down
.
2.2. THE BINOMIAL TREE MODEL 19
We calculate that the new qprobabilities are exactly 1/2 at each and every node. Now
we can work out the value of the option at the penultimate time 2 by applying the up
down formulae to the ﬁnal nodes in adjacent pairs. Figure 2.6 shows the result of the
ﬁrst two such calculations.
We can complete ﬁlling in the nodes on level 2, and then repeat the process on
level 1, and so on. At the end of this process we have the completed tree (ﬁgure 2.7).
The price of the option at time zero is 15. We can trace through our hedge, using
the formula that, at any current time, we should hedge
φ =
f
up
−f
down
s
up
−s
down
units of stock.
Figure 2.6: The option claims and claimvalues at time 2
Figure 2.7: The option claim tree
20 CHAPTER 2. DISCRETE PROCESSES
Time 0 We are given 15 for the option. We calculate φ as (25 −5)/(120 −80) = 0.5.
Buying 0.5 units of stock costs 50, so we need to borrow an additional 35.
Suppose the stock now goes up to 120
Time 1 The new φ is (40 − 10)/(140 − 100) = 0.75, so we buy another 0.25 units of
stock at its new price, taking our total borrowing to 65.
Suppose the stock goes up again to 140
Time 2 The new φ is (60 − 20)/(160 − 120) = 1, so we take our stock holding up to
1, making our debt now 100.
Finally suppose the stock goes down to 120
Time 3 The option will be in the money, and we are exactly placed to hand over one
unit of stock and receive 100 in cash to cancel our debt. (In fact, the same
would have happened if the stock had gone up to 160 instead.)
The table below shows exactly how the various processes change over time. The
portfolio strategies shown are those in force for the previous tickperiod, for instance,
φ
1
units of stock are held during the interval from i = 0 to i = 1. The option value
matches the worth of both the old and the new portfolios, for instance V
1
equals both
φ
1
S
1
+ ψ
1
and φ
2
S
1
+ ψ
2
.
Table 2.1: Option and portfolio development
Stock Option Stock Bond
Time i Last Jump Price S
i
Value V
i
Holding φ
i
Holding ψ
i
0 — 100 15 — —
1 up 120 25 0.50 35
2 up 140 40 0.75 65
3 down 120 20 1.00 100
This was the rosy scenario. What would have happened if the initial jump had
been down?
Suppose the stock goes down to 80
Time 1 This time, the new φ is (10 − 0)/(100 − 60) = 0.25. We sell half our stock
holding and reduce our debt to 15.
Suppose the stock goes up again to 100
Time 2 The next hedge is (20 −0)/(120 −80) = 0.50. We buy an extra 0.25 units of
stock and our borrowing mounts to 40.
Suppose the stock goes down again to 80
Time 3 Our stock is now worth 40, exactly canceling the debt. But the option is out
of the money, so overall we have broken even.
We note that all the process above (S, V, φ and ψ) depend on the sequence of up
down jumps. In particular, φ and ψ are random too, but depend only on the jumps
2.2. THE BINOMIAL TREE MODEL 21
Table 2.2: Option and portfolio development along a different path
Stock Option Stock Bond
Time i Last Jump Price S
i
Value V
i
Holding φ
i
Holding ψ
i
0 — 100 15 — —
1 down 80 5 0.50 35
2 up 100 10 0.25 15
3 down 80 0 0.50 40
made up to the time when you need to work them out.
xxxx; TNxxx
Exercise 2.2 Repeat the above calculations for a digital contract which pays
off 100 if the stock ends higher than it started.
The expectation result is still here. Under the probability q, the chances of each of
the ﬁnal nodes are (running from top to bottom) 1/8, 3/8, 3/8, and 1/8. The expec
tation of the claim is indeed 15 under these probabilities, but certainly not under the
model probabilities of 3/4up and 1/4down. (That gives node probabilities of 27/64,
27/64, 9/64 and 1/64, and a claim expectation of 33.75.)
Conclusions
We can sum up. The tree structure ensured that any claim provides just one possible
value for its implied derivative instrument at every node or else arbitrage intervened.
Claim led to claimvalue led to claimvalue via backwards induction until the entire
tree was ﬁlled in. Arbitrage spreads into every branch and thus across any tree.
'
&
$
%
Summary
q =
e
rδt
s
now
−s
down
s
up
−s
down
f
now
= e
−rδt
_
qf
up
+ (1 −q) f
down
_
V = f(1) = E
Q
_
B
−1
T
X
_
φ =
f
up
−f
down
s
up
−s
down
ψ = B
−1
now
(f
now
−φs
now
)
q: arbitrage probability of upjump r: interest rate in force over period
f: claim value timeprocess s: stock price process
φ: stock holding strategy B: bond price process, B
0
= 1
ψ: bond holding strategy Q: measure made up of the q
s
V : claim value at time zero X: claim payoff
δt: length of period T: time of claim payoff
22 CHAPTER 2. DISCRETE PROCESSES
Something else happened as well — each branchlet carries its own probability q
j
under which ﬁxing the value at the branchlet’s root can be given by a local expectation
operator with parameter q
j
. The cost of the local construction portfolio (φ
j
, ψ
j
) can
be written as a discounted expectation. But a string of local construction portfolios
is a global construction strategy guaranteeing a value. Thus the global discounted
expectation operator gives the value of claims on a tree as well.
2.3 Binomial representation theorem
The expectation operator is much more general and constructive an operator than its
conventional probabilistic role suggests. We can raise the apparently coincidental
ﬁnding that there exists some set of q
j
under which any derivative can be priced
by a numerically trivial discounted expectation operation to the status of a theorem.
Though it seems strangely formal here where we have the comfort of a pictured tree,
when we move to continuous models we shall be glad of any guidance — in the
continuous case intuition will often fail. And far from vanishing, the expectation
result carries across to the continuous model with ease.
It is in this spirit, then, that we derive the binomial representation theorem.
Illustrated deﬁnitions
We must start with some formal deﬁnitions of concepts we have, in many cases, al
ready met informally. There are seven separate deﬁnitions and each will be illustrated
by an example on the double forked tree with seven nodes (ﬁgure 2.8).
Figure 2.8: Tree with node numbers Figure 2.9: Tree with price process
(i) We will call the set of possible stock values, one for each node of the tree, and
their pattern of interconnections, a process S. One possible process S on our
tree is shown in ﬁgure 2.9. The random variable S
i
denotes the value of the
process at time i, for instance, S
1
is either 60 or 120 depending on whether we
are at node 2 or node 3.
2.3. BINOMIAL REPRESENTATION THEOREM 23
(a) The measure P (b) The measure Q
Figure 2.10: Measures P and Q
(ii) Separate from the process S, we will call the set of ‘probabilities’ (p
j
) or (q
j
)
a measure P or Q on the tree. The measure describes how likely any up/down
jump is at each node, represented by p
j
, the probability of moving upwards
from node j. We could choose a simple measure P with all jumps equally likely
(ﬁgure 2.10a) or a more complex measure like Q (shown in ﬁgure 2.10b).
Notice that in our formal system, we have separated two components that would
normally be seen as intimately connected parts of the same whole — the probability
of an upmove, and where the upmove is to. They may not seem too different in
character but the lesson of both the preceding sections is that this intuitive elision is
unwise. We didn’t need the real world measure P in order to ﬁnd the measure which
allowed riskfree construction. That measure was a function of S and no function
of P. The size and interrelation of upmoves affects the values of derivatives, the
probabilities of achieving them does not.
This separation of process and measure isn’t artiﬁcial — it is fundamental to ev
erything we have to do. Put crudely, the strong law failed precisely because it paid
attention to both S and P, not S alone.
(iii) A ﬁltration (F
i
) is the history of the stock up until ticktime i on the tree. The
ﬁltration starts at time zero with F
0
equal to the path consisting of the single
node 1, that is F
0
= {1}. By time 1, the ﬁltration will either be F
1
= {1, 2}
if the ﬁrst jump was down, or F
1
= {1, 3} if it was up. In full the ﬁltration
associated with each node is It thus corresponds to a particular node achieved
Table 2.3: The ﬁltration process
node 1 2 3 4 5 6 7
ﬁltration {1} {1, 2} {1, 3} {1, 2, 4} {1, 2, 5} {1, 3, 6} {1, 3, 7}
at time i. Why? Because the binomial structure ensures it — check for yourself
that there is only one path to any given node. The ﬁltration ﬁxes a history of
choices, and thus ﬁxes a node. To know where you are is the same as knowing
the ﬁltration (at least in nonrecombinant trees).
24 CHAPTER 2. DISCRETE PROCESSES
(iv) A claim X on the tree is a function of the nodes at a claim timehorizon T.
Or equivalently it is a function of the ﬁltration F
T
, thanks to the onetoone
relationship between nodes and paths. For instance, the value of the process at
time 2, S
2
, is a claim, as is the value of a call struck at 70 and the maximum
price the stock attained along its path (table 2.4).
Table 2.4: Some claims at time 2
time 2 node S
2
(S
2
−70)
+
max{S
0
, S
1
, S
2
}
7 180 110 180
6 80 10 120
5 72 2 80
4 36 0 80
The crucial difference between a claim and a process, is that the claim is only
deﬁned on the nodes at time T, while a process is deﬁned at all times up to and
including T.
Table 2.5: Conditional expectation against ﬁltration value
Expectation Filtration value Value
E
P
(S
2
F
0
) {1} (180 + 80 + 72 + 36)/4 = 92
E
P
(S
2
F
1
) {1,3}
1
2
(180 + 80) = 130
{1,2}
1
2
(72 + 36) = 54
E
P
(S
2
F
2
) {1,3,7} 180
{1,3,6} 80
{1,2,5} 72
{1,2,4} 36
(v) The conditional expectation operator E
Q
(·F
i
) extends our idea of expectation
to two parameters — a measure Q and a history F
i
. The measure Q we might
have guessed — it tells us which ‘probabilities’ to use in determining path
probability and thus the expectation. But so far we have only been interested in
taking expectations along the whole of a path from time zero, and it is useful to
take expectations from later starting points. The ﬁltration serves this purpose.
For a claim X, the quantity E
Q
(XF
i
) is the expectation of X along the latter
portion of paths which have initial segment F
i
. We regard the node reached at
time i as the new root of our tree, and take expectations of future claims from
there. This conditional expectation has an enforced dependence on the value of
the ﬁltration F
i
, and so is itself a random variable.
For each node at time i, E
Q
(XF
i
) is the expectation of X if we have already
got to that node. As an example, we take P to be the measure in ﬁgure 2.10a
and X to be the claim S
2
(table 2.5).
2.3. BINOMIAL REPRESENTATION THEOREM 25
Sensibly enough, starting at the root gives the same answer as the uncondi
tioned expectation E
P
(S
2
), whereas ‘starting’ at time 2 leaves no further time
for development, so E
P
(S
2
F
2
) = S
2
, for every possible value of the ﬁltration
F
2
.
We could also see E
P
(XF
i
) as a process in i. In the case of X = S
2
, it is
shown in ﬁgure 2.11. In this way we can convert a claim into a process, given a
measure.
Figure 2.11: Conditional expectation process E
P
(S
2
F
i
)
(vi) A previsible process φ = φ
i
is a process on the same tree whose value at any
given node at timetick i is dependent only on the history up to one timetick
earlier, F
i−1
. What can we say about a previsible process? Given the oneto
one relationship between nodes and histories on our binary tree, it is certainly
a binomial tree process in its own right, whose values are well deﬁned at each
node later than time zero. But compared to the main process S, it is known one
node in advance. It doesn’t seem to notice branches until one timestep after
they have happened. For instance a random bond price process B
i
would be
previsible, as is the delayed price process φ
i
= S
i−1
, i ≥ 1 (ﬁgure 2.12). It is not
always sensible to deﬁne the value that a previsible process has at time zero.
Figure 2.12: The previsible process S
i−1
Previsible processes will play the part of trading strategies, where we cannot
tell in advance where prices are going to go. This is an essential feature of any
26 CHAPTER 2. DISCRETE PROCESSES
model that excludes arbitrage (or insider trading).
Our ﬁnal deﬁnition is probably the most important of all — one question that we
must surely ask soon is: what is the riskfree construction measure? Is it speciﬁc to
the task in hand, or is it special in some other way as well?
(vii) A process S is a martingale with respect to a measure P and a ﬁltration (F
i
) if
E
P
(S
j
F
i
) = S
i
, for all i ≤ j.
This daunting expression needs expansion. Written out, for S to be a martingale
with respect to a measure P, it means that the future expected value at time j of
the process S under measure P (for of course our formal expectation demands a
measure, it has no meaning without one) conditional on its history up until time i
is merely the process’ value at time i. Rewritten again, that means the process
S has no drift under P, no bias up or down in its value under the expectation
operator E
P
. If the process has value 100 at some point, then its conditional
expected value under P is 100 thereafter.
Example (1). The process which constantly takes a ﬁxed value is, rather triv
ially, a martingale with respect to all possible measures.
Example (2). Our illustrative process S is actually a martingale under the mea
sure Q given in ﬁgure 2.10b. For instance E
Q
(S
1
F
0
) equals
1
3
× 120 +
2
3
×
60 = 80, and 80 is indeed the value of S
0
. Slightly harder, E
Q
(S
2
F
1
) equals
2
5
× 180 +
3
5
× 80 = 120 if the ﬁrst jump was up, which matches the value S
1
takes if the ﬁrst jump is up. The downjump case and all the others need to be
checked separately.
Example (3). The conditional expectation process N
i
= E
P
(S
2
F
i
) is a P
martingale. Because of the nature of its deﬁnition we only need to check that
E
P
(N
1
F
0
) is equal to N
1
. As this is just
1
2
×130 +
1
2
×54 = 92, it is immediate.
The last example above is a particular example of a general result.
The conditional expectation process of a claim
For any claim X, the process E
P
(XF
i
) is always a Pmartingale.
To see this to be true, we need to use the fact that
E
P
_
E
P
(XF
j
)F
i
¸
= E
P
(XF
i
), i ≤ j.
In other words, that conditioning ﬁrstly on the history up to time j and then condi
tioning on the history up to an earlier time i is the same as just conditioning originally
up to time i. This result is called the tower law.
Given the tower law, an easy check of whether a process is a Pmartingale or not
is to compare the process S
i
itself with the conditional expectation process of its
terminal value E
P
(S
T
F
i
). Only if these are identical is the process a Pmartingale.
2.3. BINOMIAL REPRESENTATION THEOREM 27
We must also take the P dependence seriously. The process S is not a martingale
on its own, it is a Pmartingale, it is a martingale with respect to the measure P. And
of course, exactly the same process can be a martingale with respect to one measure
and not to another. For instance, our illustrative process S is not a Pmartingale
(because ﬁgure 2.9 and ﬁgure 2.11 are different), but it is a Qmartingale, where Q is
given in ﬁgure 2.10b. Such a Q is called a martingale measure for S.
xxxx; TNxxx
Exercise 2.3 Check that E
Q
(S
2
F
i
) is the same as S
i
, and so prove that S is
a Qmartingale.
Binomial representation theorem
We can now write down our theorem.
Binomial representation theorem
Suppose the measure Qis such that the binomial price process S is a Qmartingale.
If N is any other Qmartingale, then there exists a previsible process φ such that
N
i
= N
0
+
i
k=1
φ
k
∆S
k
,
where ∆S
i
:= S
i
− S
i−1
is the change in S from ticktime i − 1 to i, and φ
i
is the
value of φ at the appropriate node at ticktime i.
We can get from N
0
to N
i
previsibly, with steps we know in advance. The proof
is formal but straightforward — with the work we have put in already, this kind of
manipulation should be second nature.
Figure 2.13: The branch geometry (process S on left; process N on right)
Consider a single branching from a node at ticktime i − 1 to two nodes ‘up’ and
‘down’ at ticktime i. The structure of the tree ensures that the history F
i
has two
choices beyond F
i−1
, corresponding to the up jump and down jumps respectively.
The increments over the branch of the processes S and N are
∆S
i
= S
i
−S
i−1
and ∆N
i
= N
i
−N
i−1
.
28 CHAPTER 2. DISCRETE PROCESSES
The variability that these increments contain depends on the geometry of the branch
itself (ﬁgure 2.13).
There are only two places to go, so any random variable dependent on the branch
is fully determined by its width size and a constant offset depending only on F
i−1
.
So if we want to construct one random process out of another, it will in general be
a construction based on a scaling (to match the widths) and a shift (to match the
offsets).
Consider then the scaling ﬁrst. The size of the difference between the up and
down jump values is δs
i
= s
up
−s
down
for S and δn
i
= n
up
−n
down
for N, both of these
dependent only on the ﬁltration F
i−1
. So we deﬁne φ
i
to be the ratio of these branch
widths, that is
φ
i
=
δn
i
δs
i
.
Now we can worry about the shift — the Nincrement ∆N
i
must be given by the
scaled increment φ
i
∆S
i
plus an offset k, this k again determined only by F
i−1
. That
is
∆N
i
= φ
i
∆S
i
+ k, for φ
i
and k known by F
i−1
.
But S and N are Qmartingales, that is E
Q
(∆N
i
F
i−1
) and E
Q
(∆S
i
F
i−1
) are both zero
— the increments have zero expectation conditional on the history F
i−1
. The scaling
factor is previsible, that is known by time i −1, so we also have E
Q
(φ
i
∆S
i
F
i−1
) = 0.
Thus the offset k must be zero as well (0 = 0 +k).
So the general scale and shift reduces in the case where S and N are both Q
martingales to just a scaling
∆N
i
= φ
i
∆S
i
.
And of course induction ties all these increments together to give the result we want.
Financial application
We now have a theorem; but it is a formal theorem about binomial tree processes
and measures. Nowhere in our proof do we consider portfolios of a stock and bond;
nowhere do we consider arbitrage or market implications. We go through many of the
same steps as we had to in section 2.2, but we haven’t reached a ﬁnancial conclusion.
How then can we use the binomial representation theorem for pricing?
In our binomial tree model for the market, the stock follows a binomial process
S. And if there were a measure Q which made S a martingale, we could use the
representation theorem to represent some other martingale N
i
in terms of the stock
price. The previsible φ from the theorem could act as a construction strategy. At each
point we could buy the appropriate φ
i
of the stock and we would follow the gains and
losses of the martingale N
i
.
We would be able to match the martingale step for step, starting where it starts and
ﬁnishing where it ends, wherever that might be. If the martingale ended in a claim,
than that claim would have been synthesized.
2.3. BINOMIAL REPRESENTATION THEOREM 29
Two things stand in our way, though. Firstly we have a claim X, not a martingale.
And though we would like to end up at the claim, the claim doesn’t start or end
anywhere. It isn’t a process, it’s a random variable. Secondly, we have not just
a stock but a cash bond as well. Xray vision or intuition would suggest that the
φ
i
of the binomial representation theorem is going to be a vital part of our formal
construction strategy but, to use the notation of earlier, we need a ψ
i
as well. With
each stock holding comes a bond holding.
First things ﬁrst. The claim X is a random variable but we have already seen one
trick for turning random variables into processes. Given any measure Q, we can form
the process
E
i
= E
Q
(XF
i
),
by taking conditional expectations. Even better, as we have already observed, what
ever measure Q we choose, E
i
is automatically a Qmartingale. Thus if we ﬁnd Q, a
measure under which S
i
is a Qmartingale, the appropriate E
i
is one as well.
What about the cash bond? Ultimately we will simply have to grind through the
algebra but a bit of intuition can guide us to what the answer might look like. The
cash bond B
i
represents the growth of money — $1 today is not the same as $1 at
time i, all things being equal. One dollar today is like B
i
dollars at time i. But we
would like to be in a world without the growth of money — so we could simply factor
it away.
The bond process B
i
is previsible and positive. We can assume without loss of
generality that B
0
= 1.
(i) The process B
−1
i
is another previsible process, just like B
i
itself. Call this the
discount process.
(ii) Deﬁne Z
i
:= B
−1
i
S
i
which is just as well a deﬁned process as S itself and it
subsists on the same binomial tree. Call this the discounted stock process.
(iii) The value B
−1
T
X is also a claim and because of the simple mapping from Z to
S it’s as much a claim on Z as S. Call this the discounted claim.
What, then, now?
Construction strategy
Let’s try out the trick. With Q such that Z is a Qmartingale and claim X, there
is a Qmartingale process produced from B
−1
T
X by taking conditional expectations,
E
i
= E
Q
(B
−1
T
XF
i
). By the binomial representation theorem, there is a previsible
process φ such that
E
i
= E
0
+
i
k=1
φ
k
∆Z
k
.
Now consider the following construction strategy: at ticktime i, buy the portfolio
Π
i
consisting of:
30 CHAPTER 2. DISCRETE PROCESSES
• φ
i+1
units of the stock S,
• ψ
i+1
= (E
i
−φ
i+1
B
−1
i
S
i
) units of the cash bond.
At time zero, our starting point, Π
0
is worth φ
1
S
0
+ψ
1
B
0
= E
0
= E
Q
(B
−1
T
X) — it
costs that much to create. There is also no difﬁculty in determining φ
1
or ψ
1
as φ and
ψ are previsible.
What about one tick later? We have held the portfolio safe across the period, but
its constituents have changed in value: Π
0
is now worth
φ
1
S
1
+ ψ
1
B
1
= B
1
_
E
0
+ φ
1
(B
−1
1
S
1
−B
−1
0
S
0
)
¸
,
but B
−1
1
S
1
−B
−1
0
S
0
= ∆Z
1
. Now we can use the binomial representation theorem to
simplify the expression above: at time 1, Π
0
is worth B
1
E
1
.
We are at time 1, and the construction strategy demands that we buy a new port
folio Π
1
. But the portfolio Π
1
, which we need to create at time 1, costs precisely that
amount above: B
1
E
1
, whatever actually happened to S, that is whichever ﬁltration
F
1
actually obtains.
Thus we can cash in our portfolio Π
0
to create Π
1
. And so on. At time i, portfolio
Π
i
costs B
i
E
i
to purchase, and it will change by time (i + 1) to be worth B
i+1
E
i+1
,
the cost of the next portfolio. Our construction strategy is what we might call self
ﬁnancing. And at the end of our selfﬁnancing strategy, we end up with the worth of
Π
T−1
at time T, which is B
T
B
−1
T
X. That is X, the claim we require.
Arbitrage
The price of the claim X is now obvious: it is E
Q
(B
−1
T
X) — the expected value of
the discounted claim, under the martingale measure Q for the discounted stock Z.
And it is an arbitrage price because any other price could be milked for free money
by running the (φ
i
, ψ
i
) strategy the appropriate way round to duplicate the claim. We
shouldn’t be too surprised — we are simply repeating the argument of section 2.2 in
formal guise. But our formal argument has won us an overview of the entire process
and a couple of vital slogans:
The existence of selfﬁnancing strategies
The ﬁrst slogan is that within the binomial tree model, we can produce a selfﬁnancing
(φ
i
, ψ
i
) strategy which duplicates any claim. What do we mean exactly by self
ﬁnancing? Let us deﬁne V
i
, the worth of the trading strategy at time i, to be the
opening value of the portfolio Π
i
at time i, that is V
i
= φ
i+1
S
i
+ψ
i+1
B
i
. Then a strat
egy is selfﬁnancing if the closing value of the portfolio Π
i−1
at time i is precisely
equal to V
i
. In symbols, the ‘ﬁnancing gap’ of cash that would otherwise have to be
injected into the strategy,
D
i
= V
i
−φ
i
S
i
−ψ
i
B
i
,
must be zero.
2.3. BINOMIAL REPRESENTATION THEOREM 31
Another way of representing this selfﬁnancing property comes from the changes
of the strategy value process ∆V
i
= V
i
−V
i−1
,
∆V
i
= φ
i
∆S
i
+ ψ
i
∆B
i
+ D
i
.
The gap D
i
at time i is zero if and only if the change in value of the strategy from
time i −1 to i is due only to changes in the stock and bond values alone.
Formally:
Selfﬁnancing hedging strategies
Given a binomial tree model of a market with a stock S and bond B, then (φ
i
, ψ
i
)
is a selfﬁnancing strategy to construct a claim X if:
(i) both φ and ψ are previsible;
(ii) the change in value V of the portfolio deﬁned by the strategy obeys the dif
ference equation:
∆V
i
= φ
i
∆S
i
+ ψ
i
∆B
i
where ∆S
i
:= S
i
− S
i−1
is the change in S from ticktime i − 1 to i, and
∆B
i
:= B
i
−B
i−1
is the corresponding change in B;
(iii) and is identically equal to the claim X.
Expectation of the discounted claim under the martingale measure
The second of these slogans is that the price of any derivative within the binomial tree
model is the expectation of the discounted claim under the measure Q which makes
the discounted stock a martingale.
Option price formula (discrete case)
The value at ticktime i of a claim X maturing at date T is
B
i
E
Q
(B
−1
T
XF
i
).
Why? Precisely because there is a selfﬁnancing strategy, justiﬁed by the binomial
representation theorem, which requires that amount to start off and yields the claim
without risk at T.
Uniqueness and existence of Q
And in this discrete world, we can add almost as an afterthought that for any sensible
stock process S, there will be a unique measure Qunder which B
−1
i
S
i
, the discounted
stock, is a Qmartingale.
32 CHAPTER 2. DISCRETE PROCESSES
Conclusions
We are now ﬁnished in the discrete world, we have the general theorem we require.
Any claim on a stock implies a derivative instrument tied to the underlying stock
value at any time by a construction strategy capable of providing arbitrage riches if
any market player disobeyed it. That arbitrage justiﬁed value is the expectation of
the discounted claim, but expectation under just one special measure, the measure
Q under which the discounted stock is a martingale. The real measure P which S
follows is irrelevant. The construction strategy is selfﬁnancing and generates the
claim whatever S does.
2.4 Overture to continuous models
We can, in a heuristic way, look into the continuous world with our discrete tech
niques. Without being fully rigorous yet, we could believe that a continuous model
can be approximated by a discrete time model with a very small intertick time. In
deed we can show that a natural discrete model with constant growth rate and noise
approximates a lognormal distribution under both the original measure P and the
martingale measure Q. It will even be possible to ‘derive’ the BlackScholes option
pricing formula, though its rigorous development must wait until the very end of the
next chapter.
Model with constant stock growth and noise
The model is parameterized by the intertick time δt. As that quantity gets smaller,
the model should ever more closely approximate a continuoustime model. There are
also three ﬁxed and constant parameters: the noisiness σ, the stock growth rate µ, and
the riskless interest rate r.
The cash bond B
t
has the simple form that B
t
= exp(rt), which does not depend
on the interval size.
The stock process follows the nodes of a recombinant tree, which moves from
value s at some particular node along the next up/down branch to the new value
_
_
_
s exp
_
µδt + σ
√
δt
_
if up,
s exp
_
µδt −σ
√
δt
_
if down.
The jumps are all equally likely to be up as down, that is p = 1/2 everywhere.
For a ﬁxed time t, if we set n to be the number of ticks till time t, then n = t/δt
and
S
t
= S
0
exp
_
µt + σ
√
t
_
2X
n
−n
√
n
__
,
where X
n
is the total number of the n separate jumps which were upjumps. The
random variable X
n
has the binomial distribution with mean n/2 and variance n/4,
so that (2X
n
−n)/
√
n has mean zero and variance one. By the central limit theorem,
2.4. OVERTURE TO CONTINUOUS MODELS 33
this distribution converges to that of a normal random variable with zero mean and
unit variance. So as δt gets smaller and n gets larger, the distribution of S
t
becomes
lognormal, as log S
t
is normally distributed with mean log S
0
+ µt and variance σ
2
t.
Under the martingale measure
This is what happens under the original measure P, but what goes on with Q?
Following our formula the martingale measure probability q is
q =
s exp(rδt) −s
down
s
up
−s
down
.
We can calculate that q is approximately equal to
q =
1
2
_
1 −
√
δt
_
µ +
1
2
σ
2
−r
σ
__
.
So, under Q, X
n
is still binomially distributed, but now has mean nq and variance
nq(1 −q).
Thus (2X
n
− n)/
√
n has mean −
√
t(µ +
1
2
σ
2
− r)/σ and variance asymptotically
approaching one. Again the central limit theorem gives the convergence of this to a
normal random variable with the same mean and variance exactly one. The corre
sponding S
t
is still lognormally distributed with log S
t
having mean log S
0
+(r−
1
2
σ
2
)t
and variance σ
2
t. This can be written
S
t
= S
0
exp
_
σ
√
tZ + (r −
1
2
σ
2
)t
_
,
where Z is a normal N(0, 1) under Q. We have found the marginal distribution of S
t
under the martingale measure Q.
Pricing a call option
If X is the call option maturing at date T, struck at k, with X = (S
T
− k)
+
, then its
worth at time zero is
E
Q
(B
−1
T
X) = E
Q
_
_
S
0
exp(σ
√
TZ −
1
2
σ
2
T) −k exp(−rT)
_
+
_
.
We will see in chapter three that this evaluates as
S
0
Φ
_
log
S
0
k
+
_
r +
1
2
σ
2
_
T
σ
√
T
_
−ke
−rT
Φ
_
log
S
0
k
+
_
r −
1
2
σ
2
_
T
σ
√
T
_
,
where Φ is the normal distribution function Φ(x) = Q(Z ≤ x). This is a preview of
the BlackScholes formula which we shall prove properly in the next chapter.
Chapter 3
Continuous processes
S
tock prices are not trees. The discrete trees of the previous chapter are only an
approximation to the way that prices actually move. In practice, a price can
change at any instant, rather than just at some ﬁxed ticktimes when a portfolio
can be calmly rebalanced. The binary choice of a single jump ‘up’ or ‘down’ only
becomes subtle as the ticks get closer and closer, giving the tree more and evershorter
branches. But such trees grow too complex and we stop being able to see the wood.
We shall have to start from scratch in the continuous world. The discrete models
will guide us — the intuitions gained there will be more than useful — but limiting
arguments based on letting δt tend to zero are too dangerous to be used rigorously.
We will encounter a representation theorem which establishes the basis of riskfree
construction and again it will be martingale measures that prime the expectation op
erator correctly. But processes and measures will be harder to separate intuitively —
we will need a calculus to help us. And changes in measure will affect processes in
surprising ways. We will no longer be able to proceed in full generality — we will
concentrate on Brownian motion and its relatives. If there is one overarching prin
ciple to this chapter it is that Brownian motion is sophisticated enough to produce
interesting models and simple enough to be tractable. Given the subtleties of work
ing with continuous processes, a simple calculus based on Brownian motion will be
more than enough for us.
3.1 Continuous processes
We want randomness. With our discrete stock price model we didn’t have any old
random process. We forcibly limited ourselves to a binomial tree. We started simply
and hoped (with some justiﬁcation) that complex enough market models could be
built from such humble materials. The single binomial branching was the building
block for our ‘realistic’ market. For the continuous world we need an analogous basis
— something simple and yet a reasonable starting point for realism.
What is a continuous process? Three smallscale principles guide us. Firstly, the
value can change at any time and from moment to moment. Secondly, the actual
values taken can be expressed in arbitrarily ﬁne fractions — any real number can be
34
3.1. CONTINUOUS PROCESSES 35
taken as a value. And lastly the process changes continuously — the value cannot
make instantaneous jumps. In other words, if the value changes from 1 to 1.05 it
must have passed through, albeit quickly, all the values in between.
At least as a starting point, we can insist that stock market indices or prices of
individual securities behave this way. Even though they move in a ‘sharpedged’
way, it isn’t too unrealistic to claim that they nonetheless display continuous process
behavior.
And as far back as Bachelier in 1900, who analyzed the motion of the Paris stock
exchange, people have gone further and compared the prices to one particular contin
uous process — the process followed by a randomly moving gas particle, or Brown
ian motion (ﬁgure 3.2).
Figure 3.1: UK FTA index, 196392 Figure 3.2: Brownian motion
Locally the likeness can be striking — both display the same jaggedness, and
the same similarity under scale changes — the jaggedness never smooths out as the
magniﬁcation increases. But globally, the similarity fades — ﬁgure 3.1 doesn’t look
like ﬁgure 3.2. At an intuitive level, the global structure of the stock index is different.
It grows, gets ‘noisier’ as time passes, and doesn’t go negative. Brownian motion
can’t be the whole story.
But we only want a basis — the single binomial branching didn’t look promis
ing right away. We shouldn’t run ahead of ourselves. Brownian motion will prove a
remarkably effective component to build continuous processes with — locally Brow
nian motion looks realistic.
Brownian motion
It was nearly a century after botanist Robert Brown ﬁrst observed microscopic parti
cles zigzagging under the continuous buffeting of a gas that the mathematical model
for their movements was properly developed. The ﬁrst step to the analysis of Brow
nian motion is to construct a special family of discrete binomial processes.
In other words, if X
1
, X
2
, . . . is a sequence of independent binomial random vari
ables taking values +1 or −1 with equal probability, then the value of W
n
, at the ith
step is deﬁned by:
W
n
(
i
n
) = W
n
(
i−1
n
) +
X
i
√
n
, for all i ≥ 1.
36 CHAPTER 3. CONTINUOUS PROCESSES
The ﬁrst two steps are shown in ﬁgure 3.3. What does W
n
look like as n gets large?
Instead of blowing out of control, the family portraits (ﬁgure 3.4) appear to be
settling down towards something as n increases. The moves of size 1/
√
n seem to
force some kind of convergence. Can we make a formal statement? Consider for
example, the distribution of W
n
at time 1: for a particular W
n
, there are n+1 possible
values that it can take, ranging from−
√
n to
√
n. But the distribution always has zero
mean and unit variance. (Because W
n
(1) is the sum of n IID random variables, each
with zero mean and variance 1/n.)
Figure 3.3: The ﬁrst two steps of the random walk W
n
Figure 3.4: Random walks of 16, 64, 256 and 1024 steps respectively
Moreover the central limit theorem gives us a limit for these binomial distributions
— as n gets large, the distribution of W
n
(1) tends towards the unit normal N(0, 1). In
fact, the value of W
n
(t) is the same as
W
n
(t) =
√
t
_
nt
i=1
X
i
√
nt
_
.
The distribution of the ratio in brackets tends, by the central limit theorem, to a
normal N(0, 1) random variable. And so the distribution of W
n
(t) tends to a normal
3.1. CONTINUOUS PROCESSES 37
N(0, t).
There is a formal unity underlying the family — all the marginal distributions tend
towards the same underlying normal structure.
And not just all the marginal distributions, but all the conditional marginal distri
butions as well. Each random walk W
n
has the property that its future movements
away from a particular position are independent of where that position is (and indeed
independent of its entire history of movements up to that time). Additionally such a
future displacement W
n
(s +t) −W
n
(s) is binomially distributed with zero mean and
variance t. Thus again, the central limit theorem gives us a constant limiting struc
ture, and all conditional marginals tend towards a normal distribution of the same
mean and variance.
The marginals converge, the conditional marginals converge, and the temptation
is irresistible to say that the distributions of the processes converge too. And indeed
they do, though this isn’t the place to set up the careful formal framework to make
sense of that statement. The distribution of W
n
converges, and it converges towards
Brownian motion.
Formally:
Brownian motion
The process W = (W
t
: t ≥ 0) is a PBrownian motion if and only if
(i) W
t
is continuous, and W
0
= 0,
(ii) the value of W
t
is distributed, under P, as a normal random variable N(0, t),
(iii) the increment W
s+t
− W
s
, is distributed as a normal N(0, t), under P, and is
independent of F
s
, the history of what the process did up to time s.
These are both necessary and sufﬁcient conditions for the process W to be Brow
nian motion. The last condition, though an exact echo of the behavior of the dis
crete precursors W
n
(t), is subtle. Many processes that have marginals N(0, t) are not
Brownian motion. In the continuous world, just as it was in the discrete, it is not just
the marginals (conditional on the process’ value at time zero) that count, but all the
marginals conditional on all the histories F
s
. It will in fact be the daunting task of
specifying all these that drives us to a Brownian calculus.
xxxx; TNxxx
Exercise 3.1 If Z is a normal N(0, 1), then the process X
t
=
√
tZ is con
tinuous and is marginally distributed as a normal N(0, t). Is X a Brownian
motion?
38 CHAPTER 3. CONTINUOUS PROCESSES
xxxx; TNxxx
Exercise 3.2 If W
t
and
˜
W
t
are two independent Brownian motions and ρ
is a constant between −1 and 1, then the process X
t
= ρW
t
+
_
1 −ρ
2 ˜
W
t
is continuous and has marginal distributions N(0, t). Is this X a Brownian
motion?
It is also worth noting just how odd Brownian motion really is. We won’t stop to
prove them, but here is a brief peek into the bestiary:
• Although W is continuous everywhere, it is (with probability one) differentiable
nowhere.
• Brownian motion will eventually hit any and every real value no matter how
large, or how negative. It may be a million units above the axis, but it will (with
probability one) be back down again to zero, by some later time.
• Once Brownian motion hits a value, it immediately hits it again inﬁnitely often,
and then again from time to time in the future.
• It doesn’t matter what scale you examine Brownian motion on — it looks just
the same. Brownian motion is a fractal.
Brownian motion is often also called a Wiener process, and is a (onedimensional)
Gaussian process.
Brownian motion as stock model
We had our misgivings about Brownian motion as a global model for stock behavior,
but we don’t have to use it on its own. Brownian motion wanders. It has mean zero,
whereas the stock of a company normally grows at some rate — and historically we
expect prices to rise if only because of inﬂation. But we can add in a drift artiﬁcially.
For example the process S
t
= W
t
+µt, for some constant µ reﬂecting nominal growth,
is called Brownian motion with drift.
And if it looks too noisy, or not noisy enough, we can scale the Brownian motion
by some factor: for example, S
t
= σW
t
+ µt, for a constant noise factor σ.
How are we doing? Consider the stock market data shown in ﬁgure 3.1. We could
estimate σ and µ for the best ﬁt [in this case, σ = 91.3 and µ = 37.81] and simulate a
sample path.
Not bad — the process has longterm upwards growth, as we want. But in this
particular case, we have a glitch right away. The process went negative, which we
may not want for the price of a stock of a limited liability company.
xxxx; TNxxx
Exercise 3.3 Show that, for all values of σ (σ = 0), µ, and T > 0 there is al
ways a positive probability that S
T
is negative. (Hint: consider the marginal
distribution of S
T
.)
3.2. STOCHASTIC CALCULUS 39
Figure 3.5: Brownian motion plus drift
We can though be more adventurous in shaping Brownian motion to our ends.
Consider for example, taking the exponential of our process:
X
t
= exp(σW
t
+ µt).
Now we mirror the stock market’s longterm exponential growth (and for good mea
sure we start off quietly and get noisier). Again ﬁnding a best ﬁt for σ and µ
[σ = 0.178 and µ = 0.087, a ‘noisiness’ of 17.8% and an annual drift of 8.7%] we can
simulate a sample path (ﬁgure 3.6).
Figure 3.1: UK FTA index, 196392 Figure 3.6: Exponential Brownian motion
This process is, not surprisingly well known and it is usually called exponential
Brownian motion with drift, or sometimes geometric Brownian motion with drift. It
is not the only model for stocks — and indeed we will look at others later on — but
it is simple and not that bad. (Could you tell which picture was which without the
captions?) Brownian motion can prove an effective building block.
3.2 Stochastic calculus
Shaping Brownian motion with functions may be powerful, but it brings a dangerous
complexity. Consider any smooth (differentiable) curve. Globally it can have almost
any behavior it likes, because the condition that it is differentiable does nothing to
affect it at a large scale. Suppose we zoom in though, pinning down a small section
under a microscope. In ﬁgure 3.7, we focus in on the point of a particular differ
entiable curve with xcoordinate of 1.7, increasing the magniﬁcation by a factor of
about ten each time.
40 CHAPTER 3. CONTINUOUS PROCESSES
Reading the graphs from left to right and line by line, each small box is expanded
to form the frame for the next graph. As the process continues, the graph section
becomes smoother and straighter, until eventually it is straight — it is a small straight
line.
Figure 3.7: Progressive magniﬁcation around the point 1.7
Differentiable functions, however strange their global behavior, are at heart built
from straight line segments. Newtonian calculus is the formal acknowledgement of
this.
With a Newtonian construction, we could decide to build up a family of nice func
tions by specifying howthey are locally built up out of our building block, the straight
line. We would write the change in value of a Newtonian function f over a time in
terval at t of inﬁnitesimal length dt as
df
t
= µ
t
dt,
where µ
t
is our scaling function, the slope or drift of the magniﬁed straight line at t.
Then we could explore our universe of Newtonian functions. Consider, for exam
ple:
(1) The equation df
t
= µdt, for some constant µ. What is f? That is, what does it
look like? How does it behave globally? Could we draw it? If we stick together
straight line segments of slope µ, then intuitively we just produce a straight line
of slope µ. If f
0
, for example, was equal to zero then we might guess (correctly)
that f
t
could be written in more familiar notation as f
t
= µt.
(2) The equation df
t
= tdt. Here we have a slope at time t of value t — what does
this look like? Simple integration comes to the rescue. If f
0
= 0, then we could
again pin down f
t
as f
t
=
1
2
t
2
. The going was a bit harder here, but we managed
it, and we can check it ourselves by differentiation: f
t
= t as we require.
3.2. STOCHASTIC CALCULUS 41
What about uniqueness though? In the ﬁrst example, our intuition dismissed the
possibility of another solution, but what about here? The construction metaphor
(df
t
= tdt tells us how to build f
t
, and thus given a starting place and a deterministic
building plan we ought to produce just one possible f
t
) suggests that f
t
=
1
2
t
2
is the
unique solution and indeed we can formalize this.
Uniqueness of Newtonian differentials
Two complementary forms of uniqueness operate here.
• If f
t
and
˜
f
t
are two differentiable functions agreeing at 0 (f
0
=
˜
f
0
) and they
have identical drifts (df
t
= d
˜
f
t
), then the processes are equal: (f
t
=
˜
f
t
) for all
t. In other words, f is unique given the drift µ
t
(and f
0
).
• Secondly, given a differentiable function f
t
, there is only one drift function µ
t
which satisﬁes f
t
= f
0
+
_
t
0
µ
s
ds (for all t). So µ is unique given f.
Instead of just giving the drift µ
t
directly, we might have a problem where the drift
itself depends on the current value of the function. Speciﬁcally, if the drift µ
t
equals
µ(f
t
, t), where µ(x, t) is a known function, then
df
t
= µ(f
t
, t)dt
is called an ordinary differential equation (ODE). If there is a differentiable function
f which satisﬁes it (with given f
0
), it forms a solution. There are plenty of ODEs
which have no solutions, and plenty more which do not have unique solutions. (The
uniqueness of the solution to an ODE cannot be deduced just from the uniqueness of
Newtonian differentials in the box.)
(3) The equation df
t
= f
t
dt. Now things are harder, as direct integration is not a
route to the solution. We could guess — say f
t
= e
t
— and then check by
differentiation. This solution happens to be unique for f
0
= 1.
(4) The equation df
t
= f
t
t
−2
dt. This is an example of a bad case, where solutions
need neither exist nor be unique. Given f
0
= 0, there are an inﬁnite number
of solutions, namely f
t
= a exp(−1/t) for every possible value of an arbitrary
constant a. However, for f
0
= 0, there are no solutions at all.
Perhaps our universe of Newtonian functions isn’t so benign. It is clear that though
ODEs are powerful construction tools, they are also dangerous ones. There are plenty
of ‘bad’ ODEs which we haven’t a clue how to explore.
Stochastic differentials
And if it was bad for Newtonian differentials, consider a construction procedure
based on Brownian motion. Zooming in on Brownian motion doesn’t produce a
straight line. (ﬁgure 3.8)
42 CHAPTER 3. CONTINUOUS PROCESSES
Figure 3.8: ‘Zooming in’ on Brownian motion
As before, each box is expanded by suitable horizontal and vertical scaling to
frame the next graph. The selfsimilarity of Brownian motion means that each new
graph is also a Brownian motion, and just as noisy.
But of course this selfsimilarity is ideal for a building block — we could build
global Brownian motion out of lots of local Brownian motion segments. And we
could build general randomprocesses fromsmall segments of Brownian motion (suit
ably scaled). If we built using straight line segments (suitably scaled) too, we could
include Newtonian functions as well.
A stochastic process X will have both a Newtonian term based on dt and a Brow
nian term, based on the inﬁnitesimal increment of W which we will call dW
t
. The
Brownian term of X can have a noise factor σ
t
, and so the inﬁnitesimal change of X
t
is
dX
t
= σ
t
dW
t
+ µ
t
dt.
As in the Newtonian case, the drift µ
t
can depend on the time t. But it can also be
random and depend on values that X (or indeed W) took up until t itself. And of
course, so can the noisiness σ
t
. Such processes, like X and σ, whose value at time t
can depend on the history F
t
, but not the future, are called adapted to the ﬁltration F
of the Brownian motion W.
We call σ
t
the volatility of the process X at time t and µ
t
the drift of X at t.
Stochastic processes
What does our universe look like? As with Newtonian differentials, ﬁnding this out
entails ‘integrating’ stochastic differentials in some way. We can, though, formally
deﬁne what it is to be a (continuous) stochastic process.
This deﬁnition of stochastic process (see box) is not universal, and in particular it
excludes discontinuous cases such as Poisson processes. Nevertheless it will be quite
adequate for all the models we will meet.
3.2. STOCHASTIC CALCULUS 43
The technical condition that σ and µ must be Fprevisible processes means that
they are adapted to the ﬁltration F, and that they may have some jump discontinuities.
In terms of stochastic analysis, this deﬁnes stochastic processes to be semimartingales
whose drift termis absolutely continuous. This class is closed under all the operations
used later, and all the models considered will lie within it.
And as it happens, we can provide a uniqueness result to mirror the classical setup.
Stochastic processes
A stochastic process X is a continuous process (X
t
: t ≥ 0) such that X
t
can be
written as
X
t
= X
0
+
_
t
0
σ
s
dW
s
+
_
t
0
µ
s
ds,
where σ and µ are randomFprevisible processes such that
_
t
0
(σ
2
s
+µ
s
)ds is ﬁnite
for all times t (with probability 1). The differential form of this equation can be
written
dX
t
= σ
t
dW
t
+ µ
t
dt.
Uniqueness of volatility and drift
Two complementary forms of uniqueness operate here.
• Firstly, if two processes X
t
and
˜
X
t
agree at time zero (X
0
=
˜
X
0
) and they
have identical volatility σ
t
and drift µ
t
, then the processes are equal: X
t
=
˜
X
t
for all t. In other words, X is unique given σ
t
and µ
t
(and X
0
).
• Secondly, given the process X, there is only one pair of volatility σ
t
and drift
µ
t
which satisﬁes X
t
= X
0
+
_
t
0
σ
s
dW
s
+
_
t
0
µ
s
ds (for all t). This uniqueness of
σ
t
and µ
t
given X comes from the DoobMeyer decomposition of semimartin
gales.
In the special case when σ and µ depend on W only through X
t
, such as σ
t
=
σ(X
t
, t), where σ(x, t) is some deterministic function, the equation
dX
t
= σ(X
t
, t)dW
t
+ µ(X
t
, t)dt
is called a stochastic differential equation (SDE) for X. And it will generally be
easier to write down the SDE (if it exists) for a particular X then it is to provide an
explicit solution for the SDE. As in the Newtonian case (ODEs), an SDE need not
have a solution, and if it does it might not be unique. Usage of the term SDE does
tend to spread out from this strict deﬁnition to include the stochastic differentials of
processes whose volatility and drift depends not only on X
t
and t, but also on other
events in the history F
t
.
But can we recognize the world we have created, perhaps in terms of W
t
, the
Brownian motion we have some handle on?
Partially. In the simple case, where σ and µ are both constants, meaning that X
44 CHAPTER 3. CONTINUOUS PROCESSES
has constant volatility and drift, the SDE for X is
dX
t
= σdW
t
+ µdt.
It isn’t too hard to guess what the solution to this is:
X
t
= σW
t
+ µt,
(assuming that X
0
= 0). And our meager understanding of W
t
and dW
t
at least
gives us some conﬁdence that the differential form of σW
t
is σdW
t
. As σ and µ are
independent of X, the uniqueness result could form a part of a proof that this is the
only solution.
But consider the only slightly more complex SDE (echoing the Newtonian ODE of
example (3) above),
dX
t
= X
t
(σdW
t
+ µdt).
We’re at sea.
3.3 Itˆ o calculus
Intuitive integration doesn’t carry us very far. We need tools to manipulate the dif
ferential equations, just as Newtonian calculus has the chain rule, product rule, inte
gration by parts, and so on.
How far could Newton carry us? Suppose we had some function f of Brownian
motion, say f(W
t
) = W
2
t
. Could we use a simple chain rule to produce the stochastic
differential df
t
? Under Newtonian rules, dW
2
t
would be 2W
t
dW
t
, which doesn’t look
too implausible. But we should check via integration, because
if
_
t
0
d(W
2
s
) = 2
_
t
0
W
s
dW
s
, then W
2
t
= 2
_
t
0
W
s
dW
s
.
How can we tackle
_
t
0
W
s
dW
s
? Consider dividing up the time interval [0, t] into a
partition {0, t/n, 2t/n, . . . , (n −1)t/n, t} for some n. Then we could approximate the
integral with a summation over this partition, that is
2
_
t
0
W
s
dW
s
≈ 2
n−1
i=1
W
_
it
n
_
_
W
_
(i+1)t
n
_
−W
_
it
n
_
_
.
Now something begins to worry us. The difference term inside the brackets is just
the increment of Brownian motion from one particular partition point to the next. By
property (iii) of Brownian motion, that increment is independent of the Brownian
motion up to that point, and in particular it is independent of the Brownian motion
term W(it/n). Also the increment has zero mean, which means that so too must the
product of the increment and W(it/n). So the summation consists of terms with zero
mean, forcing it to have zero mean itself.
But W
2
t
has mean t, because of the variance structure of Brownian motion, so
2W
t
dW
t
cannot be the differential of W
2
t
, because its integral doesn’t even have the
right expectation.
3.3. IT
ˆ
O CALCULUS 45
What went wrong? Consider a Taylor expansion of f(W
t
) for some smooth f:
df(W
t
) = f
(W
t
)dW
t
+
1
2
f
(W
t
)(dW
t
)
2
+
1
3!
f
(W
t
)(dW
t
)
3
+· · ·
Overfamiliar with Newtonian differentials, we assumed that (dW
t
)
2
and higher terms
were zero. But as we have observed before, Brownian motion is odd. Take (dW
t
)
2
,
given the same partitioning of [0, t] we just used: {0, t/n, 2t/n, . . . , t}. We can model
the integral of (dW
t
)
2
by the (hopefully convergent) approximation
_
t
0
(dW
t
)
2
=
n
i=1
_
W
_
ti
n
_
−W
_
t(i−1)
n
__
2
.
But if we let Z
n,i
be
Z
n,i
=
W
_
ti
n
_
−W
_
t(i−1)
n
_
_
t/n
,
then for each n, the sequence Z
n,1
, Z
n,2
, . . . is a set of IID normals N(0, 1). (Because
each increment W
_
ti
n
_
− W
_
t(i−1)
n
_
a normal N(0, t/n), independent of the ones
before it, by Brownian motion fact (iii).)
We can rewrite our approximation for
_
(dW
s
)
2
as
_
t
0
(dW
s
)
2
≈ t
n
i=1
Z
2
n,i
n
.
By the weak law of large numbers (just like the strong law but only talking about the
distribution of random variables), the distribution of the right hand side summation
converges towards the constant expectation of each Z
2
n,i
, namely 1. Thus
_
t
0
(dW
s
)
2
=
t, or in differential form (dW
t
)
2
= dt.
We can’t ignore (dW
t
)
2
; it only looks second order because of the notation. What
about (dW
t
)
3
and so on? It turns out that they are zero. (For example, E(dW
t

3
) has
size (dt)
3/2
, which is negligible compared with dt.) So Taylor gives us:
df(W
t
) = f
(W
t
)dW
t
+
1
2
f
(W
t
)dt + 0.
The formal version of this surprising departure from Newtonian differentials is the
deservedly famous Itˆ o’s formula (sometimes seen modestly as It ˆ o’s lemma).
Itˆ o’s formula
If X is a stochastic process, satisfying dX
t
= σ
t
dW
t
+µ
t
dt, and f is a deterministic
twice continuously differentiable function, then Y
t
:= f(X
t
) is also a stochastic
process and is given by
dY
t
=
_
σ
t
f
(X
t
)
_
dW
t
+
_
µ
t
f
(X
t
) +
1
2
σ
2
t
f
(X
t
)
_
dt.
Returning to our W
2
t
, we can apply It ˆ o with X = W and f(x) = x
2
and we have
d(W
2
t
) = 2W
t
dW
t
+ dt, or W
2
t
= 2
_
t
0
W
s
dW
s
+ t,
46 CHAPTER 3. CONTINUOUS PROCESSES
which at least has the right expectation.
More generally, if X is still just the Brownian motion W, then f(X) has differential
df(W
t
) = f
(W
t
)dW
t
+
1
2
f
(W
t
)dt,
as hinted above.
xxxx; TNxxx
Exercise 3.4 If X
t
= exp(W
t
), then what is dX
t
?
SDEs from processes
It ˆ o’s most immediate use is to generate SDEs from a functional expression for a pro
cess. Consider the exponential Brownian motion we set up in section 3.1:
X
t
= exp(σW
t
+ µt).
What SDE does X follow? We know we can handle the term inside the brackets but
we have to take a stochastic differential of the exponential function as well. With the
right formulation though, we can use It ˆ o’s formula.
Suppose we took Y
t
to be the process σW
t
+ µt, and f to be the exponential func
tion f(x) = e
x
. Then Y
t
is simple enough that we can write down its differential
immediately: dY
t
= σdW
t
+ µdt. But of course the X
t
we want can be written as
X
t
= f(Y
t
), so one application of It ˆ o’s formula gives us
dX
t
= σf
(Y
t
)dW
t
+
_
µf
(Y
t
) +
1
2
σ
2
f
(Y
t
)
_
dt.
The exponential function is particularly pleasant, as f
(Y
t
) = f
(Y
t
) = f(Y
t
) = X
t
, so
we can rewrite the differential as
dX
t
= X
t
_
σdW
t
+
_
µ +
1
2
σ
2
_
dt
¸
.
Here, the variable σ is sometimes called the logvolatility of the process, because it
is the volatility of the process log X
t
, and which is often abbreviated just to volatility
notwithstanding that term’s existing deﬁnition. We will also use the name logdrift
for the drift µ of log X
t
, which is different from the drift of dX
t
/X
t
above.
Processes from SDEs
Much like differentiation (easy, but its inverse can be impossible), using It ˆ o to convert
processes to SDEs is relatively straightforward. And if that were all we ever wanted
to do there would be few problems. But it isn’t — one of the key needs we have is
to go in the opposite direction and convert SDEs to processes. Or in other words, to
solve them.
In general we can’t. Most stochastic differential equations are just too difﬁcult to
solve. But a few, rare examples can be, and just like some ODEs they depend on an
3.3. IT
ˆ
O CALCULUS 47
inspired guess and then a proof that the proposed solution is an actual solution via
It ˆ o. Such a solution to an SDE is called a diffusion.
Suppose we are asked to solve the SDE
dX
t
= σX
t
dW
t
.
We need an inspired guess — so we notice that the stochastic term (σX
t
dW
t
) from
this SDE is the same as the SDE we generated via It ˆ o in the section above. Moreover,
if we choose µ to be −
1
2
σ
2
, then the drift term in the SDE would match our SDE as
well. We guess then that
X
t
= exp
_
σW
t
−
1
2
σ
2
t
_
.
What does It ˆ o tell us? That dX
t
is indeed σX
t
dW
t
, which is what we wanted. So
we have found one solution, and as it turns out, the only solution (up to constant
multiples). Soluble SDEs are scarce, and this one is special enough to have a name:
the Doleans exponential of Brownian motion.
Let’s go back then to the SDE we tripped over earlier:
dX
t
= X
t
(σdW
t
+ µdt).
We could match both drift and volatility terms for this SDE and the SDE of exp(σW
t
+
νt) if and only if we take ν to be µ −
1
2
σ
2
. So that is our guess, that
X
t
= exp
_
σW
t
+
_
µ −
1
2
σ
2
_
t
_
.
And again It ˆ o conﬁrms our intuition.
xxxx; TNxxx
Exercise 3.5 What is the solution of dX
t
= X
t
(σdW
t
+µ
t
dt), for µ
t
a general
bounded integrable function of time?
The product rule
Another Newtonian law was the product rule, that d(f
t
g
t
) = f
t
dg
t
+ g
t
df
t
. In the
stochastic world, there are two (seemingly) separate cases.
In the more signiﬁcant case, X
t
and Y
t
are adapted to the same Brownian motion
W, in that
dX
t
= σ
t
dW
t
+ µ
t
dt,
dY
t
= ρ
t
dW
t
+ ν
t
dt.
By applying It ˆ o’s formula to
1
2
_
(X
t
+ Y
t
)
2
−X
2
t
−Y
2
t
_
= X
t
Y
t
, we can see that
d(X
t
Y
t
) = X
t
dY
t
+ Y
t
dX
t
+ σ
t
ρ
t
dt.
The ﬁnal term above is actually dX
t
dY
t
(following from (dW
t
)
2
= dt) and again
marks the difference between Newtonian and It ˆ o calculus.
48 CHAPTER 3. CONTINUOUS PROCESSES
In the other case, X
t
and Y
t
are two stochastic processes adapted to two different
and independent Brownian motions, such as
dX
t
= σ
t
dW
t
+ µ
t
dt,
dY
t
= ρ
t
d
˜
W
t
+ ν
t
dt.
where σ
t
and ρ
t
are the respective volatilities of X and Y , µ
t
and ν
t
are their drifts,
and W and
˜
W are two independent Brownian motions. Here
d(X
t
Y
t
) = X
t
dY
t
+ Y
t
dX
t
,
just as in the Newtonian case.
At a deeper level these two stochastic cases can be reconciled by viewing X
t
and
Y
t
as both adapted to the twodimensional Brownian motion (W
t
,
˜
W
t
), as will be
explained in section 6.3.
xxxx; TNxxx
Exercise 3.6 Show that if B
t
is a zerovolatility process and X
t
is any
stochastic process, then
d(B
t
X
t
) = B
t
dX
t
+ X
t
dB
t
.
3.4 Change of measure — the CMG theorem
Something remains hidden from us. One of the central themes of the previous chapter
was the importance of separating process and measure. Yet we don’t seem to mention
measures in our stochastic differentials. We may have our basic tools for manipulat
ing stochastic processes, but they are a manipulation of differentials of Brownian
motion, not a manipulation of measure. We haven’t actually ignored the importance
of measure — W
t
is not strictly a Brownian motion per se, but a Brownian motion
with respect to some measure P, a PBrownian motion. And thus our stochastic
differential formulation describes the behavior of the process X with respect to the
measure P that makes the W
t
(or of course the dW
t
) a Brownian motion. But the
only tool we have seen so far gives us no clue how W
t
let alone X
t
changes as the
measure changes.
As it happens, Brownian motions change in easy and pleasant ways under changes
in measure. And thus by extension through their differentials, so do stochastic pro
cesses.
Change of measure — the RadonNikodym derivative
To get some intuitive feel for the effects of a change of measure, we should go back
for a while to discrete processes. Consider a simple twostep random walk:
3.4. CHANGE OF MEASURE — THE CMG THEOREM 49
Figure 3.9: Twostep recombinant tree
Table 3.1: Path probabilities
Path Probability
{0, 1, 2} p
1
p
2
=: π
1
{0, 1, 0} p
1
(1 −p
2
) =: π
2
{0, −1, 0} (1 −p
1
)p
3
=: π
3
{0, −1, −2} (1 −p
1
)(1 −p
3
) =: π
4
To get from time 0 to time 2, we can follow four possible paths {0, 1, 2}, {0, 1, 0},
{0, −1, 0}, {0, −1, −2}. Suppose we speciﬁed the probability of taking these paths:
We could view this mapping of paths to path probabilities as encoding the measure
P. If we knew π
1
, π
2
, π
3
and π
4
, then (as long as all of them are strictly between 0 and
1) we know p
1
, p
2
and p
3
. Thus if we represent our process with a nonrecombining
tree, we can label each of the paths at the end with the πinformation encoding the
measure.
Figure 3.10: Tree with path probabilities marked
Now suppose we had a different measure Q with probabilities q
1
, q
2
and q
3
. Again
we can code this up with path probabilities, say π
1
, π
2
, π
3
and π
4
. And again if each
π
is strictly between 0 and 1, π
1
, π
2
, π
3
and π
4
uniquely decides Q.
50 CHAPTER 3. CONTINUOUS PROCESSES
And with this encoding, there is a very natural way of encoding the differences
between P and Q, giving some idea of how to distort P so as to produce Q. If we
form the ratio π
i
/π
i
for each path i, we write the mapping of paths to this ratio as
dQ
dP
.
This random variable (random because it depends on the path) is called the Radon
Nikodym derivative of Q with respect to P up to time 2.
Figure 3.11: Tree with RadonNikodym derivative marked
From
dQ
dP
we can derive Q fromP. How? If we have P, then we have π
1
, π
2
, . . . , π
4
,
and
dQ
dP
gives us the ratios π
i
/π
i
, so we have π
1
, π
2
, . . . , π
4
. And thus Q.
What about p
i
or q
i
being zero or one? Two things happen — ﬁrstly it can become
impossible to back out the p
i
fromthe π
i
. Consider if p
1
is zero then both π
1
and π
2
are
zero and so information about p
2
is lost. But then of course, the paths corresponding
to π
1
and π
2
are both impossible (probability zero), so in some sense p
2
really isn’t
relevant. If we restrict ourselves to only providing π
i
for possible paths, then we can
recover the corresponding p’s.
The second problem has a similar ﬂavor but is more serious. Suppose one of the
p’s is zero, but none of the q’s are. Then at least one π
i
will be zero when none of
the π
i
are. Not all the ratios π
i
/π
i
will be well deﬁned, and thus
dQ
dP
can’t exist. We
could suppress those paths which had path probability zero, but now we have lost
something. Those paths may have been Pimpossible but they are Qpossible. If we
throw them away, then we have lost information about Q just where it is relevant
— paths which are Qpossible. Somehow we can’t deﬁne
dQ
dP
if Q allows something
which P doesn’t. And of course vice versa.
This is important enough to formalize.
Equivalence
Two measures P and Q are equivalent if they operate on the same sample space
and agree on what is possible. Formally, if A is any event in the sample space,
P(A) > 0 ⇐⇒Q(A) > 0.
In other words, if A is possible under P then it is possible under Q, and if A is
impossible under P then it is also impossible under Q. And vice versa.
3.4. CHANGE OF MEASURE — THE CMG THEOREM 51
We can only meaningfully deﬁne
dQ
dP
and
dP
dQ
if P and Q are equivalent, and then
only where paths are Ppossible. But of course if paths are Pimpossible then we
know how Q acts on those paths — if Q is equivalent to P then they are Qimpossible
as well.
Thus two measures P and Q must be equivalent before they will have Radon
Nikodym derivatives
dQ
dP
and
dP
dQ
.
Expectation and
dQ
dP
While we are still working with discrete processes, we should stock up on some
facts about expectation and the RadonNikodym derivative. One of the reasons for
deﬁning it was the efﬁcient coding it represented. Everything we needed to know
about Q could be extracted from P and
dQ
dP
.
Consider then a claim X known by time 2 on our discrete twostep process. The
claim X is a random variable, or in other words a mapping from paths to values —
we can let x
i
denote the value the claim takes if path i is followed. So the expectation
of X with respect to P is given by
E
P
(X) =
i
π
i
x
i
,
where i ranges over all four possible paths. And the expectation of X with respect to
Q is
E
Q
(X) =
i
π
i
x
i
=
i
π
_
π
i
π
i
x
i
_
= E
P
_
dQ
dP
X
_
.
Just like X,
dQ
dP
is a random variable which we can take the expectation of. And the
conversion from Q to P is pleasingly simple: E
Q
(X) = E
P
_
dQ
dP
X
_
.
Attractive though this is, it represents just one simple case:
dQ
dP
is deﬁned with a
particular time horizon in mind — the ends of the paths, in this case T = 2. We
speciﬁed X at this time and we only wanted an unconditioned expectation. In formal
terms, the result we derived was
E
Q
(X
T
F
0
) = E
P
_
dQ
dP
X
T
¸
¸
¸
¸
F
0
_
,
where T is the time horizon for
dQ
dP
and X
T
is known at time T. What about E
Q
(X
t
F
s
)
for t not equal to T and s not equal to zero? We need somehow to know
dQ
dP
not just
for the ends of paths but everywhere —
dQ
dP
is a random variable, but we would like a
process.
RadonNikodym process
We can do this by letting the time horizon vary, and setting ζ
t
to be the Radon
Nikodym derivative taken up to the horizon t. That is, ζ
t
is the RadonNikodym
derivative
dQ
dP
but only following paths up to time t, and only looking at the ratio of
52 CHAPTER 3. CONTINUOUS PROCESSES
probabilities up to that time. For instance, at time 1, the possible paths are {0, 1}
and {0, −1} and the derivative ζ
1
has values on them of q
1
/p
1
and (1 − q
1
)/(1 − p
1
)
respectively. At time zero, the derivative process is just 1, as the only ‘path’ is the
point {0} which has probability 1 under both P and Q. Concretely, we can ﬁll in ζ
t
on our tree in terms of the p’s and q’s (ﬁgure 3.12).
Figure 3.12: Tree with ζ
t
process marked (¯ p
i
= 1 −p
i
, ¯ q
i
= 1 −q
i
)
In fact there is another expression for ζ
t
as the conditional expectation of the T
horizon RadonNikodym derivative,
ζ
t
= E
P
_
dQ
dP
¸
¸
¸
¸
F
t
_
,
for every t less than or equal to the horizon T.
xxxx; TNxxx
Exercise 3.7 Prove that this equation holds for t = 0, 1, 2.
We can see that the expectation with respect to P unpicks the
dQ
dP
in just the right
way. The process ζ
t
represents just what we wanted — an idea of the amount of
change of measure so far up to time t along the current path. If we wanted to know
E
Q
(X
t
) it would be E
P
(ζ
t
X
t
), where X
t
is a claim known at time t. If we want to
know E
Q
(X
t
F
s
) then we need the amount of change of measure from time s to time
t — which is just ζ
t
/ζ
s
. That is, the change up to time t with the change up to time s
removed. In other words
E
Q
(X
t
F
s
) = ζ
−1
s
E
P
(ζ
t
X
t
F
s
).
xxxx; TNxxx
Exercise 3.8 Prove this on the tree.
3.4. CHANGE OF MEASURE — THE CMG THEOREM 53
'
&
$
%
RadonNikodym summary
Given P and Q equivalent measures and a time horizon T. we can deﬁne a random
variable
dQ
dP
deﬁned on Ppossible paths, taking positive real values, such that
(i) E
Q
(X
T
) = E
P
_
dQ
dP
X
T
_
, for all claims X
T
knowable by time T.
(ii) E
Q
(X
t
F
s
) = ζ
−1
s
E
P
(ζ
t
X
t
F
s
), s ≤ t ≤ T,
where ζ
t
is the process E
P
_
dQ
dP
¸
¸
¸ F
t
_
.
Change of measure — the continuous RadonNikodym derivative
What now? To deﬁne a measure for Brownian motion it seems we have to be able
to write down the likelihood of every possible path the process can take, ranging
across not only a continuousvalued state space but also a continuousvalued time
line. Standard probability theory gives some clue to the technology required, if we
were content merely to represent the marginal distributions for the process at each
time. Despite the continuous nature of the state space, we know that we can express
likelihoods in terms of a probability density function.
For example, the measure P on the real numbers, corresponding to a normal
N(0, 1) random variable X, can be represented via the density f
P
(x), where
f
P
(x) =
1
√
2π
e
−
1
2
x
2
.
In some loose sense, f
P
(x) represents the relative likelihood of the event {X = x}
occurring. Or, less informally the probability that X lies between x and x + dx is
approximately f
P
(x)dx . In exact terms, the probability that X takes a value in some
subset A of the reals is
P(x ∈ A) =
_
A
1
√
2π
e
−
1
2
x
2
dx.
For example, the chance of X being in the interval [0, 1] is the integral of the density
over the interval,
_
1
0
f
P
(x)dx, which has value 0.3413.
But marginal distributions aren’t enough — a single marginal distribution won’t
capture the nature of the process (we can see that clearly even on a discrete tree).
Nor will all the marginal distributions for each time t. We need nothing less than all
the marginal distributions at each time t conditional on every history F
s
for all times
s < t. We need to capture the idea of a likelihood of a path in the continuous case, by
means of some conceptual handle on a particular path speciﬁed for all times t < T.
One approach is to specify a path if not for all times before the horizon T, then at
least for some arbitrarily large yet still ﬁnite set of times {t
0
= 0, t
1
, . . . , t
n−1
, t
n
= T}.
Consider then, the set of paths which go through the points {x
1
, . . . , x
n
} at times
{t
1
, . . . , t
n
}. If there were just one time t
1
and one point x
1
, then we could write
54 CHAPTER 3. CONTINUOUS PROCESSES
down the likelihood of such a path. We could use the probability density function of
W
t
1
, f
1
P
(x), which is the density function of a normal N(0, t
1
), or
f
1
P
(x) =
1
√
2πt
1
exp
_
−
x
2
2t
1
_
.
And if we can do this for one time t
1
, then we can for ﬁnitely many t
i
. All we
require is the joint likelihood function f
n
P
(x
1
, . . . , x
n
) for the process taking values
{x
1
, . . . , x
n
} at times {t
1
, . . . , t
n
}.
Figure 3.13: Two Brownian motions agreeing on the set {t
1
, t
2
, t
3
}
Joint likelihood function for Brownian motion
If we take t
0
and x
0
to be zero, and write ∆x
i
for x
i
− x
i−1
and ∆t
i
= t
i
− t
i−1
,
then given the third condition of Brownian motion that increments ∆W
i
= W(t
i
) −
W(t
i−1
) are mutually independent, we can write down
f
n
P
(x
1
, . . . , x
n
) =
n
i=1
1
√
2π∆t
i
exp
_
−
(∆x
i
)
2
2∆t
i
_
.
So we can write down a likelihood function corresponding to the measure P for a
process on a ﬁnite set of times. And in the continuous limit, we have a handle on the
measure P for a continuous process. If A is some subset of R
n
, then the Pprobability
that the random nvector (W
t
, . . . , W
t
n
) is in A is exactly the integral over A of the
likelihood function f
n
P
.
Just as the measure P can be approached through a limiting time mesh, so can
the RadonNikodym derivative
dQ
dP
. The event of paths agreeing with ω on the mesh,
A = {ω
: W
t
i
(ω
) = W
t
i
(ω), i = 1, . . . , n}, gets smaller and smaller till it is just the
single pointset {ω}. The RadonNikodym derivative can be thought of as the limit
dQ
dP
(ω) = lim
A→{ω}
Q(A)
P(A)
.
3.4. CHANGE OF MEASURE — THE CMG THEOREM 55
RadonNikodym derivative — continuous version
Suppose P and Q are equivalent measures. Given a path ω, for every ordered time
mesh {t
1
, . . . , t
n
} (with t
n
= T), we deﬁne x
i
to be W
t
i
(ω), and then the derivative
dQ
dP
up to time T is deﬁned to be the limit of the likelihood ratios
dQ
dP
(ω) = lim
n→∞
f
n
Q
(x
1
, . . . , x
n
)
f
n
P
(x
1
, . . . , x
n
)
,
as the mesh becomes dense in the interval [0, T].
This continuoustime derivative
dQ
dP
still satisﬁes the results that
(i) E
Q
(X
T
) = E
P
_
dQ
dP
X
T
_
,
(ii) E
Q
(X
t
F
s
) = ζ
−1
s
E
P
(ζ
t
X
t
F
s
), s ≤ t ≤ T,
where ζ
t
is the process E
P
_
dQ
dP
¸
¸
¸ F
t
_
, and X
t
is any process adapted to the history
F
t
.
Simple changes of measure — Brownian motion plus constant drift
We have the mechanics of change of measure but still no clue about what change of
measure does in the continuous world. Suppose, for example, we had a PBrownian
motion W
t
. What does W
t
look like under an equivalent measure Q — is it still
recognizably Brownian motion or something quite different?
Foresight can provide one simple example. Consider W
t
a PBrownian motion,
then (out of nowhere) deﬁne Q to be a measure equivalent to P via
dQ
dP
= exp
_
−γW
T
−
1
2
γ
2
T
_
,
for some time horizon T. What does W
t
look like with respect to Q?
One place to start, and it is just a start, is to look at the marginal of W
T
under
Q. We need to ﬁnd the likelihood function of W
T
with respect to Q, or something
equivalent. One useful trick is to look at momentgenerating functions:
Identifying normals
A random variable X is a normal N(µ, σ
2
) under a measure P if and only if
E
P
(exp(θX)) = exp
_
θµ +
1
2
θ
2
σ
2
_
, for all real θ.
To calculate E
Q
(exp(θW
T
)), we can use fact (i) of the RadonNikodym derivative
summary, which tells us that it is the same as the Pexpectation E
P
_
dQ
dP
exp(θW
T
)
_
.
This equals
E
P
_
exp(−γW
T
−
1
2
γ
2
T + θW
T
)
_
= exp
_
−
1
2
γ
2
T +
1
2
(θ −γ)
2
T
_
,
because W
T
is a normal N(0, T) with respect to P.
56 CHAPTER 3. CONTINUOUS PROCESSES
Simplifying the algebra, we have
E
Q
(exp(θW
T
)) = exp
_
−θγT +
1
2
θ
2
T
_
,
which is the momentgenerating function of a normal N(−γT, T). Thus the marginal
distribution of W
T
, under Q, is also a normal with variance T but with mean −γT.
What about W
t
for t less than T? The marginal distribution of WT is what we
would expect if W
t
under Q were a Brownian motion plus a constant drift −γ. Of
course, a lot of other process also have a marginal normal N(−γT, T) distribution at
time T, but it would be an elegant result if the sole effect of changing fromP to Q via
dQ
dP
= exp
_
−γW
T
−
1
2
γ
2
T
_
were just to punch in a drift of −γ.
And so it is. The process W
t
is a Brownian motion with respect to P and Brownian
motion with constant drift −γ under Q. Using our two results about
dQ
dP
, we can prove
the three conditions for
˜
W
t
= W
t
+ γ
t
to be QBrownian motion:
(i)
˜
W
t
is continuous and
˜
W
0
= 0;
(ii)
˜
W
t
is a normal N(0, t) under Q;
(iii)
˜
W
t+s
−
˜
W
t
is a normal N(0, t) independent of F
s
The ﬁrst of these is true and (ii) and (iii) can be reexpressed as
(ii)
E
Q
_
exp(θ
˜
W
t
)
_
= exp
_
1
2
θ
2
t
_
;
(iii)
E
Q
_
exp
_
θ (
˜
W
t+s
−
˜
W
t
)
_¸
¸
¸ F
s
_
= exp
_
1
2
θ
2
t
_
.
xxxx; TNxxx
Exercise 3.9 Show that (ii)
and (iii)
are equivalent to (ii) and (iii) respec
tively, and prove them using the change of measure process
ζ
t
= E
P
_
dQ
dP
¸
¸
¸
¸
F
t
_
.
That both W
t
and
˜
W
t
are Brownian motion, albeit with respect to different mea
sures, seems paradoxical. But switching from P to Q just changes the relative likeli
hood of a particular path being chosen. For example, W might follow a path which
drifts downwards for a time at a rate of about −γ. Although that path is Punlikely,
it is Ppossible. Under Q, on the other hand, such a path is much more likely, and
the chances are that is what we see. But it still could be just improbable Brownian
motion behavior.
We can see this in the RadonNikodym derivative
dQ
dP
which is large when W
T
is very negative, and small when W
T
is closer to zero or positive. This is just the
consequence of the common sense thought that paths which end up negative are
more likely under Q (Brownian motion plus downward drift) than they are under P
(driftless Brownian motion). Correspondingly, paths which ﬁnish near or above zero
are less likely under Q than P.
3.4. CHANGE OF MEASURE — THE CMG THEOREM 57
CameronMartinGirsanov
So this one change of measure just changed a vanilla Brownian motion into one with
drift — nothing else. And of course, drift is one of the elements of our stochastic
differential form of processes. In fact all that measure changes on Brownian mo
tion can do is to change the drift. All the processes that we are interested in are
representable as instantaneous differentials made up of some amount of Brownian
motion and some amount of drift. The mapping of stochastic differentials under P to
stochastic differentials under Q is both natural and pleasing.
This is what our theorem provides.
CameronMartinGirsanov theorem
If W
t
is a PBrownian motion and γ
t
is an Fprevisible process satisfying the
boundedness condition E
P
exp
_
1
2
_
T
0
γ
2
t
dt
_
< ∞, then there exists a measure Q
such that
(i) Q is equivalent to P
(ii)
dQ
dP
= exp
_
−
_
T
0
γ
t
dW
t
−
1
2
_
T
0
γ
2
t
dt
_
(iii)
˜
W
t
= W
t
+
_
t
0
γ
s
ds is a QBrownian motion.
In other words, W
t
is a drifting QBrownian motion with drift −γ
t
at time t.
Within constraints, if we want to turn a PBrownian motion W
t
into a Brownian
motion with some speciﬁed drift −γ
t
, then there’s a Q which does it.
Within limits, drift is measure and measure drift.
Conversely to the theorem,
CameronMartinGirsanov converse
If W
t
is a PBrownian motion, and Q is a measure equivalent to P, then there exists
an Fprevisible process γ
t
such that
˜
W
t
= W
t
+
_
t
0
γ
s
ds
is a QBrownian motion. That is, W
t
plus drift γ
t
is QBrownian motion. Ad
ditionally the RadonNikodym derivative of Q with respect to P (at time T) is
exp
_
−
_
T
0
γ
t
dW
t
−
1
2
_
T
0
γ
2
t
dt
_
.
CMG and stochastic differentials
The CMG theorem applies to Brownian motion, but all our processes are disguised
Brownian motions at heart. Now we can see the rewards of our Brownian calculus
instantly — CMG becomes a powerful tool for controlling the drift of any process.
58 CHAPTER 3. CONTINUOUS PROCESSES
Suppose that X is a stochastic process with increment
dX
t
= σ
t
dW
t
+ µ
t
dt,
where W is a PBrownian motion. Suppose we want to ﬁnd if there is a measure Q
such that the drift of process X under Q is ν
t
dt instead of µ
t
dt. As a ﬁrst step, dX can
be rewritten as
dX
t
= σ
t
_
dW
t
+
_
µ
t
−ν
t
σ
t
_
dt
_
+ ν
t
dt.
If we set γ
t
to be (µ
t
− ν
t
)/σ
t
, and if γ then satisﬁes the CMG growth condition
E
P
exp
_
1
2
_
T
0
γ
2
t
dt
_
< ∞ then indeed there is a new measure Q such that
˜
W
t
:=
W
t
+
_
t
0
(µ
s
−ν
s
)/σ
s
ds is a QBrownian motion.
But this means that the differential of X under Q is
dX
t
= σ
t
d
˜
W
t
+ ν
t
dt,
where
˜
W is a QBrownian motion — which gives X the drift ν
t
we wanted.
We can also set limits on the changes that changing to an equivalent measure can
wreak on a process. Since the change of measure can only change the Brownian
motion to a Brownian motion plus drift, the volatility of the process must remain the
same.
Examples — changes of measure
1. Let X
t
be the drifting Brownian process σW
t
+ µt, where W is a PBrownian
motion and σ and µ are both constant. Then using CMG with γ
t
= µ/σ, there
exists an equivalent measure Q under which
˜
W
t
= W
t
+ (µ/σ)t and
˜
W is a Q
Brownian motion up to time T. Then X
t
= σ
˜
W
t
, which is (scaled) QBrownian
motion.
The measures also give rise to different expectations. For example, E
P
(X
2
t
)
equals µ
2
t
2
+ σ
2
t, but E
Q
(X
2
t
) = σ
2
t.
2. Let X
t
be the exponential Brownian motion with SDE
dX
t
= X
t
(σdW
t
+ µdt),
where W is PBrownian motion. Can we change measure so that X has the new
SDE
dX
t
= X
t
(σdW
t
+ νdt),
for some arbitrary constant drift ν?
Using CMG with γ
t
= (µ − ν)/σ, there is indeed a measure Q under which
˜
W
t
= W
t
+ (µ −ν)t/σ is a QBrownian motion. Then X does have the SDE
dX
t
= X
t
(σd
˜
W
t
+ νdt),
where
˜
W is a QBrownian motion.
3.5. MARTINGALE REPRESENTATION THEOREM 59
3.5 Martingale representation theorem
We can solve some SDEs with It ˆ o; we can see how SDEs change as measure changes.
But central to answering our pricing question in chapter two was the concept of a
measure with respect to which the process was expected to stay the same, the mar
tingale measure for our discrete trees. The price of derivatives turned out to be an
expectation under this measure, and the construction of this expectation even showed
us the trading strategy required to justify this price. And so it is here.
First the description again:
Martingales
A stochastic process M
t
is a martingale with respect to a measure P if and only if
(i) E
P
(M
t
) < ∞, for all t
(ii) E
P
(M
t
F
s
) = M
s
, for all s ≤ t.
The ﬁrst condition is merely a technical sweetener, it is the second that carries
the weight. A martingale measure is one which makes the expected future value
conditional on its present value and past history merely its present value. It isn’t
expected to drift upwards or downwards.
Some examples:
(1) Trivially, the constant process S
t
= c (for all t) is a martingale with respect to any
measure: E
P
(S
t
F
s
) = c = S
s
, for all s ≤ t, and for any measure P.
(2) Less trivially, PBrownian motion is a Pmartingale. Intuitively this makes sense
— Brownian motion doesn’t move consistently up or down, it’s as likely to do
either. But we should get into the habit of checking this formally: we need
E
P
(W
t
F
s
) = W
s
. Of course we have that the increment W
t
−W
s
, is independent
of F
s
and distributed as a normal N(0, t − s), so that E
P
(W
t
− W
s
F
s
) = 0. This
yields the result, as
E
P
(W
t
F
s
) = E
P
(W
s
F
s
) +E
P
(W
t
−W
s
F
s
) = W
s
+ 0
(3) For any claim X depending only on events up to time T, the process N
t
=
E
P
(XF
t
) is a Pmartingale (assuming only the technical constraint E
P
(X) <
∞).
Example (3) is an elegant little trick for producing martingales — and as we shall
see (and have already seen in chapter two) central to pricing derivatives. First why?
Convince yourself that N
t
= E
P
(XF
t
) is a welldeﬁned process — the ﬁrst stage of
the alchemy is the introduction of a time line into the random variable X. Now for
N
t
to be a Pmartingale, we require E
P
(N
t
F
s
) = N
s
, but for this we merely need to
be satisﬁed that
E
P
_
E
P
(XF
t
)F
s
_
= E
P
(XF
s
).
60 CHAPTER 3. CONTINUOUS PROCESSES
That is, that conditioning ﬁrstly on information up to time t and then on information
up to time s is just the same as conditioning up to time s to begin with. This property
of conditional expectation is the tower law.
xxxx; TNxxx
Exercise 3.10 Showthat the process X
t
= W
t
+γ
t
, where W
t
is a PBrownian
motion, is a Pmartingale if and only if γ = 0.
Representation
In chapter two, we had a binomial representation theorem — if M
t
and N
t
are both P
martingales then they share more than just the name — locally they can only differ by
a scaling, by the size of the opening of each particular branching. We could represent
changes in N
t
by scaled changes in the other nontrivial Pmartingale. Thus N
t
itself
can be represented by the scaled sum of these changes.
In the continuous world:
Martingale representation theorem
Suppose that M
t
is a Qmartingale process, whose volatility σ
t
satisﬁes the ad
ditional condition that it is (with probability one) always nonzero. Then if
N
t
is any other Qmartingale, there exists an Fprevisible process φ such that
_
T
0
φ
2
t
σ
2
t
dt < ∞with probability one, and N can be written as
N
t
= N
0
+
_
t
0
φ
s
dM
s
.
Further φ is (essentially) unique.
This is virtually identical to the earlier result, with summation replaced by an
integral. As we are getting used to, the move to a continuous process extracts a for
mal technical penalty. In this case, the Qmartingale’s volatility must be positive with
probability 1 — but otherwise our chapter two result has carried across unchanged. If
there is a measure Q under which M
t
is a Qmartingale, then any other Qmartingale
can be represented in terms of M
t
. The process φ
t
is simply the ratio of their respec
tive volatilities.
Driftlessness
We need just one more tool. Thrown into the discussion of martingales was the
intuitive description of a martingale as neither drifting up or drifting down. We have,
though, a technical deﬁnition of drift via our stochastic differential formulation, An
obvious question springs to mind: are stochastic processes with no drift term always
martingales, and vice versa can martingales always be represented as just σ
t
dW
t
for
some Fprevisible volatility process σ
t
?
Nearly.
3.6. CONSTRUCTION STRATEGIES 61
One way round we can do for ourselves with the martingale representation theo
rem. If a process X
t
is a Pmartingale then with W
t
a PBrownian motion, we have
an Fprevisible process φ
t
such that
X
t
= X
0
+
_
t
0
φ
s
dW
s
.
This is just the integral form of the increment dX
t
= φ
t
dW
t
, which has no drift term.
The other way round is true (up to a technical constraint), but harder. For refer
ence:
A collector’s guide to martingales
If X is a stochastic process with volatility σ
t
(that is dX
t
= σ
t
dW
t
+ µ
t
dt) which
satisﬁes the technical condition E
_
_
_
T
0
σ
2
s
ds
_
1
2
_
< ∞, then
X is a martingale ⇐⇒X is driftless (µ
t
≡ 0).
If the technical condition fails, a driftless process may not be a martingale. Such
processes are called local martingales.
Exponential martingales
The technical constraint can be tiresome. For example, take the (driftless) SDE for an
exponential process dX
t
= σ
t
X
t
dW
t
. The condition (in this case, E
_
_
_
T
0
σ
2
s
X
2
s
ds
_
1
2
_
<
∞) is difﬁcult to check, but for these speciﬁc exponential examples, a better (more
practical) test is:
A collector’s guide to exponential martingales
If dX
t
= σ
t
X
t
dW
t
, for some Fprevisible process σ
t
, then
E
_
exp
_
1
2
_
T
0
σ
2
s
ds
__
< ∞=⇒X is a martingale.
We also note that the solution to the SDE is X
t
= X
0
exp
_
_
t
0
σ
s
dW
s
−
1
2
_
t
0
σ
2
s
ds
_
.
xxxx; TNxxx
Exercise 3.11 If σ
t
is a bounded function of both time and sample path,
show that dX
t
= σ
t
X
t
dW
t
is a Pmartingale.
3.6 Construction strategies
We have the mathematical tools — It ˆ o, CameronMartinGirsanov, and the martin
gale representation theorem — now we need some idea of how to hook them into
62 CHAPTER 3. CONTINUOUS PROCESSES
a ﬁnancial model. In the simplest models, BlackScholes for example, we’ll have a
market consisting of one random security and a riskless cash account bond; and with
this comes the idea of a portfolio.
The portfolio (φ, ψ)
A portfolio is a pair of processes φ
t
and ψ
t
which describe respectively the number
of units of security and of the bond which we hold at time t. The processes can
take positive or negative values (we’ll allow unlimited shortselling of the stock or
bond). The security component of the portfolio φ should be Fprevisible: depend
ing only on information up to time t but not t itself.
There is an intuitive way to think about previsibility. If φ were leftcontinuous (that
is, φ
s
tends to φ
t
as s tends upwards to t from below) then φ would be previsible. If
φ were only rightcontinuous (that is φ
s
tends to φ
t
only as s tends downwards to t
from above), then φ need not be.
Selfﬁnancing strategies
With the idea of a portfolio comes the idea of a strategy. The description (φ
t
, ψ
t
)
is a dynamic strategy detailing the amount of each component to be held at each
instant. And one particularly interesting set of strategies or portfolios are those that
are ﬁnancially selfcontained or selfﬁnancing.
A portfolio is selfﬁnancing if and only if the change in its value only depends
on the change of the asset prices. In the discrete framework this was captured via a
difference equation, and in the continuous case it is equivalent to an SDE.
What SDE?
With stock price S
t
and bond price B
t
, the value, V
t
, of a portfolio (φ
t
, ψ
t
) at time t
is given by V
t
= φ
t
S
t
+ ψ
t
B
t
. At the next time instant, two things happen: the old
portfolio changes value because S
t
and B
t
have changed price; and the old portfolio
has to be adjusted to give a new portfolio as instructed by the trading strategy (φ, ψ).
If the cost of the adjustment is perfectly matched by the proﬁts or losses made by
the portfolio then no extra money is required from outside — the portfolio is self
ﬁnancing.
In our discrete language, we had the difference equation
∆V
i
= φ
i
∆S
i
+ ψ
i
∆B
i
.
In continuous time, we get a stochastic differential equation:
Selfﬁnancing property
If (φ
t
, ψ
t
) is a portfolio with stock price S
t
and bond price B
t
, then
(φ
t
, ψ
t
) is selfﬁnancing ⇐⇒dV
t
= φ
t
dS
t
+ ψ
t
dB
t
.
3.6. CONSTRUCTION STRATEGIES 63
Suppose the stock price S
t
is given by a simple Brownian motion W
t
(so S
t
= W
t
for all t), and the bond price B
t
is constant (B
t
= 1 for all t). What kind of portfolios
are selfﬁnancing?
(1) Suppose φ
t
= ψ
t
= 1 for all t. If we hold a unit of stock and a unit of bond
for all time without change, then the value of the portfolio (V
t
= W
t
+ 1) may
ﬂuctuate, but it will all be due to ﬂuctuation of the stock. Intuitively, no extra
money is needed to come in to uphold the (φ
t
, ψ
t
) strategy and none comes out
— this (φ
t
, ψ
t
) portfolio ought to be selfﬁnancing. Checking this formally, V
t
=
W
t
+1 implies that dV
t
= dW
t
which is the same as φ
t
dS
t
+ψ
t
dB
t
, as we required
(remembering that dB
t
= 0).
(2) Suppose φ
t
= 2W
t
and ψ
t
= −t −W
2
t
. Here (φ
t
, ψ
t
) is a portfolio, φ
t
is previsible,
and the value V
t
= φ
t
S
t
+ψ
t
B
t
= W
2
t
−t. By It ˆ o’s formula, dV
t
= 2W
t
dW
t
which
is identical to φ
t
dS
t
+ ψ
t
dB
t
as required.
xxxx; TNxxx
Exercise 3.12 Verify the It ˆ o claim in (2) above (which also shows that W
2
t
−t
is a martingale).
Surprising though it seems: holding as many units of stock as twice its current
price, though a rollercoaster strategy, is exactly offset by the stock proﬁts and the
changing bond holding of −(t +W
2
t
). The (φ
t
, ψ
t
) strategy could (in a perfect market)
be followed to our heart’s content without further funding.
The second example should convince us that being selfﬁnancing is not an auto
matic property of a portfolio. The It ˆ o check worked, but it could easily have failed if
ψ
t
had been different — the (φ
t
, ψ
t
) strategy would have required injections or forced
outﬂows of cash. Every time we claim a portfolio is selfﬁnancing we have to turn
the handle on It ˆ o’s formula to check the SDE.
Trading strategies
Now we can deﬁne a replicating strategy for a claim:
Replicating strategy
Suppose we are in a market of a riskless bond B and a risky security S with volatil
ity σ
t
and a claim X on events up to time T.
A replicating strategy for X is a selfﬁnancing portfolio (φ, ψ) such that
_
T
0
σ
2
t
φ
2
t
dt < ∞and V
T
= φ
T
S
T
+ ψ
T
B
T
= X.
Why should we care about replicating strategies? For the same reason as we
wanted them in the discrete market models. The claim X gives the value of some
derivative which we need to pay off at time T. We want a price if there is one, as of
now, given a model for S and B.
64 CHAPTER 3. CONTINUOUS PROCESSES
If there is a replicating strategy (φ
t
, ψ
t
) , then the price of X at time t must be
V
t
= φ
t
S
t
+ ψ
t
B
t
. (And speciﬁcally, the price at time zero is V
0
= φ
0
S
0
+ ψ
0
B
0
) If
it were lower, a market player could buy one unit of the derivative at time t and sell
φ
t
units of S and ψ
t
units of B against it, continuing to be short (φ, ψ) until time T.
Because (φ, ψ) is selfﬁnancing and the portfolio is worth X at time T guaranteed,
the bought derivative and sold portfolio would safely cancel at time T, and no extra
money is required between times t and T. The proﬁt created by the mismatch at time
t can be banked there and then without risk. And, as usual with arbitrage, one unit
could have been many; no risk means no fear.
And of course if the derivative price had been higher than V
t
, then we could have
sold the derivative and bought the selfﬁnancing (φ, ψ) to the same effect. Replicating
strategies, if they exist, tie down the price of the claim X not just at payoff but
everywhere.
We can lay out a battle plan. We deﬁne a market model with a stock price process
complex enough to satisfy our need for realism. Then, using whatever tools we have
to hand we ﬁnd replicating strategies for all useful claims X. And if we can, we can
price derivatives in the model. The rest of the book consists of upping the stakes in
complexity of models and of claims.
3.7 BlackScholes model
We need a model to cut our teeth on. We have the tools and we’ve seen the overall
approach at the end of chapter two. So taking the stock model of section 3.1, we
will use the CameronMartinGirsanov theorem (section 3.4) to change it into a mar
tingale, and then use the martingale representation theorem (section 3.5) to create a
replicating strategy for each claim. It ˆ o will oil the works.
The model
Our ﬁrst model — basic BlackScholes
We will posit the existence of a deterministic r, µ and σ such that the bond price
B
t
and the stock price follow
B
t
= exp(rt),
S
t
= S
0
exp(σW
t
+ µt),
where r is the riskless interest rate, σ is the stock volatility and µ is the stock drift.
There are no transaction costs and both instruments are freely and instantaneously
tradable either long or short at the price quoted.
We need a model for the behavior of the stock — simple enough that we actu
ally can ﬁnd replicating strategies but not so simple that we can’t bring ourselves to
believe in it as a model of the real world.
Following in Black and Scholes’ footsteps, our market will consist of a riskless
3.7. BLACKSCHOLES MODEL 65
constantinterest rate cash bond and a risky tradable stock following an exponential
Brownian motion.
As we’ve seen in section 3.1, it is at least a plausible match to the real world. And
as we shall see here, it is quite hard enough to start with.
Zero interest rates
If there’s one parameter that throws up a smokescreen around a ﬁrst run at an analysis
of the BlackScholes model, it’s the interest rate r. The problems it causes are more
tedious than fatal — as we’ll see soon, the tools we have are powerful enough to
cope. But we’ll temporarily simplify things, and set r to be zero.
So now we begin. For an arbitrary claim X, knowable by some horizon time T,
we want to see if we can ﬁnd a replicating strategy (φ
t
, ψ
t
).
Finding a replicating strategy
We shall follow a threestep process outlined in this box here.
'
&
$
%
Three steps to replication
(1) Find a measure Q under which S
t
is a martingale.
(2) Form the process E
t
= E
Q
(XF
t
).
(3) Find a previsible process φ
t
, such that dE
t
= φ
t
dS
t
.
The tools described earlier on will be essential to do this. We shall use the
CameronMartinGirsanov theorem (section 3.4) for the ﬁrst step and the martingale
representation theorem (section 3.5) for the third one.
Step one
For two different reasons — ﬁrstly we need to apply the CameronMartinGirsanov
theorem, and secondly we need to be able to tell if S
t
is a Qmartingale for a given Q
— we want to ﬁnd an SDE for S
t
.
The stock follows an exponential Brownian motion, S
t
= exp(σW
t
+ µ
t
), so the
logarithm of the stock price, Y
t
= log(S
t
), follows a simple drifting Brownian motion
Y
t
= σW
t
+µ
t
. Thus the SDE for Y
t
is easy to write down: dY
t
= σdW
t
+µd
t
. But, of
course, It ˆ o makes it possible to write down the SDE for S
t
= exp(Y
t
) as
dS
t
= σS
t
dW
t
+
_
µ +
1
2
σ
2
_
S
t
dt.
In order for S
t
to be a martingale, the ﬁrst thing to do is to lull the drift in this
SDE. If we let γ
t
be a process with constant value γ =
_
µ +
1
2
σ
2
_
/σ, then the CM
G theorem says that there is a measure Q such that
˜
W
t
= W
t
+ γt is QBrownian
66 CHAPTER 3. CONTINUOUS PROCESSES
motion. (The technical boundedness condition is satisﬁed because γ
t
is constant.)
Substituting in, the SDE is now
dS
t
= σS
t
d
˜
W
t
.
No drift term, thus S
t
could be a Qmartingale. The exponential martingales box
(section 3.5) contains a condition in terms of σ for S
t
to be a martingale under Q.
As σ is constant, the condition holds which means that S
t
must be a Qmartingale.
Consequently, Q is the martingale measure for S
t
.
Step two
Given Q, we can convert X into a process by forming E
t
= E
Q
(XF
t
). This is, as we
have already discussed in example (3) of section 3.5, a Qmartingale.
Step three
Since there is a Q, under which both E
t
and S
t
are Qmartingales, we can invoke
the martingale representation theorem. There exists a previsible process φ
t
which
constructs E
t
= E
Q
(XF
t
) out of S
t
. (To use the theorem, we need to check that the
volatility of S
t
is always positive, but this is true because the volatility is just σS
t
, and
both σ and S
t
are always positive.) Formally:
E
t
= E
Q
(XF
t
) = E
Q
(X) +
_
t
0
φ
s
dS
s
,
or, of course, dE
t
= φ
t
dS
t
. So the martingale representation theorem tells us an
important fact: given a Q that makes S
t
a Qmartingale with positive volatility, dE
t
=
φ
t
dS
t
for some φ
t
.
We need a replicating strategy (φ
t
, ψ
t
), and it’s tempting to believe that we have
got one half of it. So we should try it, setting ψ
t
to be the only thing it can be, given
that we want the portfolio to be worth E
t
for all t.
Replicating strategy
Our strategy is to:
• hold φ
t
units of stock at time t and
• hold ψ
t
= E
t
−φ
t
S
t
units of the bond at time t.
Is it selfﬁnancing? The value of the portfolio at time t is
V
t
= φ
t
S
t
+ ψ
t
B
t
= E
t
,
because the bond B
t
is constantly equal to 1. Thus dV
t
= dE
t
, but of course dE
t
=
φ
t
dS
t
, from the martingale representation theorem.
Since dB
t
is zero, we have the selfﬁnancing condition we want, namely dV
t
=
φ
t
dS
t
+ ψ
t
dB
t
.
3.7. BLACKSCHOLES MODEL 67
Since the terminal value of the strategy V
T
is E
T
= X, we have a replicating strat
egy for X — which means there is an arbitrage price for X at all times. Speciﬁcally
there is an arbitrage price for X at time zero — the value of the (φ
t
, ψ
t
) portfolio
at time zero, which makes the price E
0
, or E
Q
(X). In other words, the price of the
claim X is its expected value under the measure that makes the stock process S
t
a
martingale.
It is worth pausing to let a few surprises sink in. The ﬁrst is just the fact that there
are replicating strategies for arbitrary claims. The model that we have chosen isn’t too
unrealistic — it has the right kind of behavior and a healthy degree of randomness.
So we might expect to fail in our search for replicating strategies. It is after all
particularly odd that despite the lack of knowledge about the claim’s eventual value,
we can nevertheless trade in the market in such a way that we always produce it.
The second surprise, and just as important, is that the price of the derivative has
such a simple expression — the expected value of the claim. It is the easiest thing to
forget that this is not the expectation of the claim with respect to the real measure of
S
t
, which is the measure that makes it an exponential Brownian motion with drift µ
and volatility σ. All that expectation could give us would be a longterm average of
the claim’s payout. And though that could be a useful thing to know in order to judge
whether punting with the derivative is worthwhile in the long run, it doesn’t give a
price. There is a replicating strategy and thus an arbitrage price for the claim. And
arbitrage always wins out.
The price happens to be an expectation, but not the expectation in a traditional
statistics sense. It could only be the expectation if quite by chance the drift µ we
believe in for the stock were exactly and precisely right to make S
t
a martingale in
the ﬁrst place (µ = −
1
2
σ
2
).
The third surprise is the simplicity of the process S
t
under its martingale measure.
If we actually want to crank the handle and calculate derivative prices for a particular
claim, we have to be able to calculate the expected value of the claim under the mar
tingale measure Q. Since the claim depends on S
t
, this normally involves calculating
the expected value, under Q, of some function of the values of S
t
up to t = T. If S
t
were an unpleasant process under Q, then this task could be unpleasant too. But S
t
is
also an exponential Brownian motion under Q. If we solve the SDE, then
S
t
= exp
_
σ
˜
W
t
−
1
2
σ
2
t
_
,
and we ﬁnd that S
t
has the same constant volatility σ and a new but also constant
drift of −
1
2
σ
2
. So if we felt that S
t
was tractable under its original measure, it is also
tractable under the martingale measure.
Nonzero interest rates
Now we can bring the interest rate r back in again. What happens if r is nonzero?
We can’t just ignore it. Suppose we did, and considered a forward contract with claim
S
T
−k for some price k. We already know that the k which gives the forward contract
68 CHAPTER 3. CONTINUOUS PROCESSES
a zero value at time zero is k = S
0
e
rT
. The arbitrage to produce this is easy to ﬁgure
out. But our rule, when r was zero, of simply taking the expected value of the claim
under the martingale measure for S
t
cannot work. In fact,
E
Q
_
S
T
−S
0
e
rT
_
= S
0
_
1 −e
rT
_
= 0.
Even discounting the claim won’t help in this case. So our rule of ﬁnding a measure
which makes St into a martingale only holds true when r is zero. When r is not zero,
the inexorable growth of cash gets in the way.
So we take a guess. If the growth of cash is annoying, simply remove it by dis
counting everything. We call B
−1
t
the discount process, and form a discounted stock
Z
t
= B
−1
t
S
t
and a discounted claim B
−1
T
X.
In this discounted world, we could be forgiven for thinking that r was zero again.
So maybe our analysis will work again. Of course, this is all just heuristic justiﬁca
tion, and the proof is only in the doing. If we can’t ﬁnd a replicating strategy then,
attractive as our guess is, it is also wrong.
Fortunately, we can. Focusing on our discounted stock process Z
t
, it is not too
hard to write down an SDE
dZ
t
= Z
t
_
σdW
t
+ (µ −r +
1
2
σ
2
)dt
¸
.
xxxx; TNxxx
Exercise 3.13 Prove it.
Step one
To make Z
t
into a martingale, we can invoke CMG just as before, only now to
introduce a drift of (µ − r +
1
2
σ
2
)/σ to the underlying Brownian motion. So there
exists (another) Q equivalent to the original measure P and a QBrownian motion
˜
W
t
such that
dZ
t
= σZ
t
d
˜
W
t
.
So Z
t
, under Q, is driftless and a martingale.
Step two
We need a process which hits the discounted claim and is also a Qmartingale. And,
as before, conditional expectation provides it, namely by forming the process E
t
=
E
Q
(B
−1
T
XF
t
).
Step three
The discounted stock price Z
t
is a Qmartingale; and so is the conditional expectation
process of the discounted claim E
t
. Thus the martingale representation theorem gives
us a previsible φ
t
such that dE
t
= φ
t
dZ
t
.
3.7. BLACKSCHOLES MODEL 69
We want to hit the real claim with amounts of the real stock, but in our shadow dis
counted world we can hit the discounted claim by holding φ
t
units of the discounted
stock. So just as a guess, let us try φ
t
out in the real world as well.
What about the bond holding? The bond holding in the discounted world is ψ
t
=
E
t
−φ
t
Z
t
, so we can try that in the real world too. Some reassurance comes from the
fact that at time T we will be holding φ
T
units of the stock and ψ
T
units of the bond
which will be worth φ
T
S
T
+ ψ
T
B
T
= B
T
E
T
= X.
So our replicating strategy is to
• hold φ
t
units of the stock at time t, and
• hold ψ
t
= E
t
−φ
t
Z
t
units of the bond.
Are we right? The value V
t
of the portfolio (φ
t
, ψ
t
) is given by V
t
= φ
t
S
t
+ψ
t
B
t
=
B
t
E
t
. Thus following exercise 3.6, we can write dV
t
as
dV
t
= B
t
dE
t
+ E
t
dB
t
.
But dE
t
is φ
t
dZ
t
(our fact from the martingale representation theorem), and so dV
t
=
φ
t
B
t
dZ
t
+ E
t
dB
t
. A bit of rearrangement tells us that E
t
= φ
t
Z
t
+ ψ
t
, and thus
dV
t
= φ
t
B
t
dZ
t
+ (φ
t
Z
t
+ ψ
t
)dB
t
= φ
t
(B
t
dZ
t
+ Z
t
dB
t
) + ψ
t
dB
t
.
But, from exercise 3.6 again, d(B
t
Z
t
) = B
t
dZ
t
+Z
t
dB
t
, and since S
t
= B
t
Z
t
, we have
dV
t
= φ
t
dS
t
+ ψ
t
dB
t
.
That is, (φ
t
, ψ
t
) is selfﬁnancing.
Selfﬁnancing strategies
A portfolio strategy (φ
t
, ψ
t
) of holdings in a stock S
t
and a nonvolatile cash bond
B
t
has value V
t
= φ
t
S
t
+ ψ
t
B
t
and discounted value E
t
= φ
t
Z
t
+ ψ
t
, where Z is
the discounted stock process Z
t
= B
−1
t
− S
t
. Then the strategy is selfﬁnancing if
either
dV
t
= φ
t
dS
t
+ ψ
t
dB
t
,
or equivalently dE
t
= φ
t
dZ
t
.
A strategy is selfﬁnancing if changes in its value are due only to changes in the as
sets’ values, or equivalently if changes in its discounted value are due only to changes
in the discounted values of the assets.
Since we know that V
T
= X, then we have proved that (φ
t
, ψ
t
) is a replicating
strategy for X. Our guesses came good.
70 CHAPTER 3. CONTINUOUS PROCESSES
'
&
$
%
Summary
Suppose we have a BlackScholes model for a continuously tradable stock and
bond, that is assuming the existence of a constant r, µ and σ such that their re
spective prices can be represented as S
t
= S
0
exp(σWt + µt) and B
t
= exp(rt).
Then all integrable claims X, knowable by some time horizon T, have associated
replicating strategies (φ
t
, ψ
t
). In addition, the arbitrage price of such a claim X is
given by
V
t
= B
t
E
Q
(B
−1
T
XF
t
) = e
−r(T−t)
E
Q
(XF
t
),
The important measure Q is not the measure which makes the stock a martingale,
but the measure that makes the discounted stock a martingale. And the arbitrage
price of the claim is the expectation under Q of the discounted claim.
So when interest rates are nonzero, what are the new rules? They are just dis
counted versions of the old rules:
'
&
$
%
Three steps to replication (discounted case)
(1) Find a measure Q under which the discounted stock price Z
t
is a martingale.
(2) Form the process E
t
= E
Q
(B
−1
T
XF
t
).
(3) Find a previsible process φ
t
, such that dE
t
= φ
t
dZ
t
.
Call options
We should price something. Following Black and Scholes, we’ll price a call option —
the right but not the obligation to buy a unit of stock for a predetermined amount at a
particular exercise date, say T. If we let this predetermined amount be k (in ﬁnancial
terms, the strike of the option), then in formal notation, our claim is max(S
T
− k, 0).
Or in more convenient notation, (S
T
−k)
+
.
First we should ﬁnd V
0
, the value of the replicating strategy (and thus the option)
at time zero. Our formula tells us that this is given by
e
−rT
E
Q
_
(S
T
−k)
+
_
,
where Q is the martingale measure for B
−1
t
−S
t
.
But how do we ﬁnd this? The ﬁrst thing to notice is the simplicity of the claim.
The value (S
T
−k)
+
only depends on the stock price at one point in time — namely
the expiry time, T. So to ﬁnd the expectation of this claim we need only ﬁnd the
marginal distribution of S
T
under Q.
And to do that, we can look at the process for S
t
written in terms of the Q
Brownian motion
˜
W
t
. Since d(log S
t
) = σd
˜
W
t
+ (r −
1
2
σ
2
)dt, if we denote the stock
3.8. BLACKSCHOLES IN ACTION 71
price at time zero, S
0
, by s, we have that log S
t
= log s + σ
˜
W
t
+ (r −
1
2
σ
2
)t and thus
S
t
= s exp
_
σ
˜
W
t
+ (r −
1
2
σ
2
)t
_
.
So the marginal distribution for S
T
is given by s times the exponential of normal
with mean (r−
1
2
σ
2
)T and variance σ
2
T. Thus if we let Z be a normal N(−
1
2
σ
2
T, σ
2
T),
we can write S
T
as se
(Z+rT)
and thus the claimas the expectation e
−rT
E
_
_
se
(Z+rT)
−k
_
+
_
,
which equals
1
√
2πσ
2
T
_
∞
log(k/s)−rT
_
se
x
−ke
−rT
_
exp
_
−
_
x +
1
2
σ
2
T
_
2
2σ
2
T
_
dx
This integral can be decomposed by a change of variables into a couple of standard
cumulative normal integrals. If we use the notation Φ(x) to denote (2π)
−
1
2
_
x
−∞
exp
_
−y
2
/2
_
dy
the probability that a normal N(0, 1) has value less than x, then we can calculate that
V
0
= V (s, T), where
BlackScholes formula
V (s, T) = sΦ
_
log
s
k
+
_
r +
1
2
σ
2
_
T
σ
√
T
_
−ke
−rT
Φ
_
log
s
k
+
_
r −
1
2
σ
2
_
T
σ
√
T
_
.
This is the BlackScholes formula for pricing European call options. (Put options,
the right to sell a unit of stock for k, can be priced as a call less a forward — putcall
parity.)
xxxx; TNxxx
Exercise 3.14 Find the change of variable and thus prove the BlackScholes
formula.
3.8 BlackScholes in action
If a stock has a constant volatility of 18% and constant drift of 8%, with continuously
compounded interest rates constant at 6%, what is the value of an option to buy the
stock for $25 in two years time, given a current stock price of $20?
The description ﬁts the BlackScholes conditions. Thus using s = 20, k = 25,
σ = 0.18, r = 0.06, and t = 2, we can calculate V
0
as $1.221.
xxxx; TNxxx
Exercise 3.15 What information about the drift was required?
Price dependence
For values of the current stock price s much smaller than the exercise price k, the
value of the formula itself gets small, signifying that the option is out of the money
72 CHAPTER 3. CONTINUOUS PROCESSES
and unlikely to recover in time. Conversely, for values of s much greater than k, the
option loses most of its optionality, and becomes a forward. Correspondingly the
option price is approximately s −ke
−rT
which is the current value of a stock forward
struck at price k for time T.
Time dependence
As the time to maturity T gets smaller, the chances of the price moving much more
decreases and the option value gets closer and closer to the claim value taken at the
current price, (s −k)
+
.
For larger times, however, the option value gets larger. An option with almost
inﬁnite time to maturity would have value approaching s, as the cost now of price k
is almost zero. It can be seen in ﬁgure 3.14 that as the time to expiration gets closer
to zero, the curve gets closer to the option shape (s −k)
+
.
Figure 3.14: Option price against stock price for times 3, 1, and 0.3. Exercise price k = $1, interest rate r = 0,
volatility σ = 1.
Volatility dependence
All else being equal, the option is worth more the more volatile the stock is. At one
extreme, if σ is very small, the option resembles a riskless bond and is just worth
(s −ke
−rT
)
+
, which is the value of the corresponding forward if the option will be in
the money and is zero otherwise. At the other extreme, if σ is very large, the option
is worth s.
American options
Sometimes an option has more optionality about it than just choosing between two
alternatives at the maturity date. American options are the most wellknown exam
ples of such derivatives, giving the right to, say, purchase a unit of stock for a strike
price k at any time up to and including the expiration date T, rather than only at that
date. The buyer of the option then has to make decisions from moment to moment to
decide when and if to call the option.
The buyer of an American call has the choice when to stop, and that choice can
only use price information up to the present moment. Such a (random) time is called
3.8. BLACKSCHOLES IN ACTION 73
a stopping time. Following a strategy which will result in exercising the option at the
stopping time T, the corresponding payoff is
(S
τ
−k)
+
at time τ.
If the option issuer knew in advance which stopping time the investor will use, the
cost at time zero of hedging that payoff is
E
Q
_
e
−rτ
(S
τ
−k)
+
_
.
As we do not know which τ will be used, we have to prepare for the worst possible
case, and charge the maximum value (maximized over all possible stopping strate
gies),
V
0
= sup
τ
E
Q
_
e
−rτ
(S
τ
−k)
+
_
.
Pricing derivatives with optionality
In general, if the option purchaser has a set of options A, and receives a payoff X
a
at time T, after choosing a in A, then the option issuer should charge
V
0
= sup
a∈A
E
Q
_
e
−rT
X
a
_
for it. If the purchaser does not exercise the option optimally, then the issuer’s
hedge will produce a surplus by date T.
That hedge in full
Returning to the original European option, one thing that would be useful to know
would be the actual replicating strategy required, that is, to actually ﬁnd out how
much stock would be required at each point of time to artiﬁcially construct the deriva
tive.
The amount of stock, φ
t
, comes from the martingale representation theorem, but
unfortunately, the theorem merely states that φ
t
exists. However the martingale rep
resentation theorem, at heart, tells us that the reason that the discounted claim can be
built from the discounted stock is that, being martingales under the same measure,
one is locally just a scaled version of the other. The process φ
t
is merely the ratio of
volatilities. Thus, intuitively, if we looked at the ratio of the change in the value of
the option caused by a move in the stock price and the change in the stock price used,
this ought to be something like φ
t
. And if we have a restricted enough claim where
the only input required from the ﬁltration for pricing the claim is the stock price at the
current moment, and moreover that the functional relation implied by this between
the value of the claim and the current stock price is smooth, then we could guess that
the partial derivative of the option value with respect to the stock price is the φ
t
we
want.
74 CHAPTER 3. CONTINUOUS PROCESSES
And so it is. For the oftenencountered case where the claim depends only on the
terminal value, the option value is a wellbehaved function of the current stock price.
Suppose the derivative X is a function of the terminal value of the stock price, so that
X = f(S
T
) for some function f(s). Then the following is true.
Terminal value pricing
If the derivative X equals f(S
T
), for some f, then in the value of the derivative at
time t is equal to V
t
= V (S
t
, t), where V (s, t) is given by the formula
V (s, t) = exp
_
−r(T −t)
_
E
Q
_
f(S
T
)S
t
= s
_
And then the trading strategy is given by φ
t
=
∂V
∂s
(S
t
, t).
Why? Consider dV
t
, the inﬁnitesimal change in the value of the option. Remem
bering that dS
t
= σS
t
d
˜
W
t
+ rS
t
dt, then It ˆ o gives us
dV
t
= d (V (S
t
, t)) =
_
σS
t
∂V
∂s
_
d
˜
W
t
+
_
rS
t
∂V
∂s
+
1
2
σ
2
S
2
t
∂
2
V
∂s
2
+
∂V
∂t
_
dt.
But we also know that dV
t
= φ
t
dS
t
+ ψ
t
dB
t
from the selfﬁnancing condition. And
since dB
t
= rB
t
dt , we have
dV
t
= (σS
t
φ
t
) d
˜
W
t
+ (rS
t
φ
t
+ rψ
t
B
t
) dt.
But SDE representations are unique — so the volatility terms must match, giving
φ
t
=
∂V
∂s
. The amount of stock in the replicating portfolio at any stage is the derivative
of the option price with respect to the stock price.
Using this substitution for φ
t
and the fact that V
t
= S
t
φ
t
+ψ
t
B
t
, we can also match
the drift terms of the two SDEs to get a partial differential equation for V as
1
2
σ
2
s
2
∂
2
V
∂s
2
+ rs
∂V
∂s
−rV +
∂V
∂t
= 0.
Notoriously, this PDE, coupled with the boundary condition that V (s, T) must
equal f(s), gives another way of solving the pricing equation.
Explicit BlackScholes hedge
The call option is a terminal value claim, as described earlier, and so we can ﬁnd
an expression for the hedge itself. The amount of stock held is the derivative of the
value function with respect to stock price. In symbols
φ
t
=
∂V
∂s
(S
t
, T −t) = Φ
_
log
S
t
k
+
_
r +
1
2
σ
2
_
(T −t)
σ
√
T −t
_
.
Because φ is always between zero and one, we need only ever have a bounded long
position in the stock. Also the value of the bond holding at any time is
B
t
ψ
t
= −ke
−r(T−t)
Φ
_
log
S
t
k
+
_
r −
1
2
σ
2
_
(T −t)
σ
√
T −t
_
,
3.8. BLACKSCHOLES IN ACTION 75
which, although always a borrowing, is bounded by the exercise price k.
There are two possibilities as the time approaches maturity. If the option is out of
the money, that is the stock price is less than the exercise price, then both the bond
and the stock holding go to zero, reﬂecting the increasing worthlessness of the option.
Alternatively, if the price stays above the exercise value, then the stock holding grows
to one unit and the value of the bond to −k. This combination exactly balances the
now certain demand for a unit of stock in return for cash amount k.
Example — hedging in continuous time
This can be seen operating in practice. Below are two possible realizations of a stock
price which starts at $10. Both are exponential Brownian motions with volatility 20%
and growth drift of 15%.
(a) Stock price (A) (b) Stock price (B)
Figure 3.15: Stock price (A) and (B)
Let us price an option on this stock, to buy it at time T = 1 for the strike price of
k = $12, assuming interest rates are 5%. We can calculate both the evolving worth of
the option V
t
and the amount of stock to be held, φ
t
, to hedge the contract.
In the case (A), these processes are shown in ﬁgure 3.16.
(a) Option value (A) (b) Stock hedge (A)
Figure 3.16: Option value and Stock hedge of (A)
As time progresses, the option becomes in the money and the option value moves
like the stock price. Also the hedge gets closer and closer to one, signifying that the
option will be exercised.
In the case (B), these processes are shown in Figure 3.17.
76 CHAPTER 3. CONTINUOUS PROCESSES
(a) Option value (B) (b) Stock hedge (B)
Figure 3.17: Option value and Stock hedge of (B)
This time the option is not exercised and both the value of it and the hedge go to
zero over time.
xxxx; TNxxx
Exercise 3.16 A stock has current price $10 and moves as an exponen
tial Brownian motion with upward drift of 15% a year (continuously com
pounded) and volatility of 20% a year. Current interest rates are constant at
5%. What is the value of an option on the stock for $12 in a year’s time?
xxxx; TNxxx
Exercise 3.17 For the same stock, what is the value of a derivative which
pays off $1 if the stock price is more than $10 in a year’s time?
Conclusions
Even with a respectable stochastic model for the stock, we can replicate any claim.
Not something we had any right to expect. The replicating portfolio has a value
given by the expected discounted claim, with respect to a measure which makes the
discounted stock a martingale. Moreover, changing to the martingale measure has a
remarkably simple effect on the process S
t
— only the drift changes, to another con
stant value. The stock remains an exponential Brownian motion; even the volatility
σ stays the same.
These three surprises conspire to make the result look easier to get at than perhaps
it really is. Something subtle and beautiful really is going on under all the formalism
and the result only serves to obscure it. Before we push on, stop and admire the
view.
Chapter 4
Pricing market securities
T
he BlackScholes model we have seen so far has a simple mathematical side
but it has an even simpler ﬁnancial side. The asset we considered was a stock
which could be held without additional cost or beneﬁt and was freely tradable
at the price quoted. Even leaving aside the issues of transaction costs and illiquid
ity, not much of the ﬁnancial market is like that. Even vanilla products — foreign
exchange, equities and bonds — don’t actually ﬁt the simple asset class we devised.
Foreign exchange involves two assets which pay interest, equities pay dividends, and
bonds pay coupons.
Just retreading the same mathematics for each of these will be enough to keep us
busy. The sophistication we have to peddle now is ﬁnancial.
4.1 Foreign exchange
In the foreign exchange market, like the stock market, holding the basic asset, cur
rency, is a risky business. The dollar value of, say, one pound sterling varies from
moment to moment just as a US stock does. And with this risk comes demand for
derivatives: claims based on the future value of one unit of currency in terms of
another.
Forwards
Consider, though, a forward transaction: a dollar investor wanting to agree the cost
in dollars of one pound at some future date T. As with stocks, the replicating strategy
to guarantee the forward claim is static. We buy pounds now and sell dollars against
them. But cash in both currencies attracts interest. And just as in the simple Black
Scholes model, our cash holding wasn’t cash but a cash bond, so our cash holdings
here will be cash bonds as well.
Let’s make things concrete. Suppose the constant dollar interest rate is r, the
sterling interest rate is u, and C
0
dollars buy a pound now. Consider the following
static replicating strategy. At time t we
• own e
−uT
units of sterling cash bonds, and
77
78 CHAPTER 4. PRICING MARKET SECURITIES
• go short C
0
e
−uT
units of dollar cash bonds.
At time zero the portfolio has nil value, and at time T the sterling holding will be
one pound as required and the dollar short holding will be C
0
e
(r−u)T
— the forward
price we require.
Contrast this with the stock forward price S
0
e
rT
. We must be careful in extending
our simple model to foreign exchange — both instruments now make payments. And
that makes a difference.
BlackScholes currency model
There are three instruments and processes to model — two local currency cash bonds
and the exchange rate itself. Following the mathematical simplicity of BlackScholes,
our market will be:
BlackScholes currency model
We let B
t
be the dollar cash bond, D
t
its sterling counterpart, and C
t
be the dollar
worth of one pound. Then our model is
Dollar bond B
t
= e
rt
,
Sterling bond D
t
= e
ut
,
Exchange rate C
t
= C
0
exp(σW
t
+ µt),
for some W
t
a PBrownian motion and constants r, u, σ and µ.
The dollar investor
The underlying ﬁnance dictates that there are two tradables available to the dollar in
vestor. One is uncomplicated — the dollar bond is straightforwardly a dollar tradable
much as the cash bond was in the basic account of BlackScholes. But the other is
not.
We would like to think of the stochastic process C
t
, the exchange rate, as a trad
able but it isn’t. The process C
t
represents the dollar value of one pound sterling, but
sterling cash isn’t a tradable instrument in our market. To hold cash naked would be
to set up an arbitrage against the cash bond — to put it another way, the existence of
the sterling cash bond D
t
sets an interest rate for sterling cash by arbitrage, and that
rate is u not zero.
On the other hand, D
t
by itself isn’t a dollar tradable either — it is the price of a
tradable instrument, but it’s a sterling price.
Fortunately, the product of the two S
t
= C
t
D
t
is a dollar tradable. The dollar
investor can hold sterling cash bonds, and the dollar value of the holding will be
given by the translation of the sterling price D
t
into dollars, that is by multiplication
by C
t
.
Translation, then, yields two processes, B
t
and S
t
, which mirror the basic Black
Scholes set up.
4.1. FOREIGN EXCHANGE 79
'
&
$
%
Three steps to replication (foreign exchange)
(1) Find a measure Q under which the sterling bond discounted by the dollar bond
Z
t
= B
−1
t
S
t
= B
−1
t
C
t
D
t
, is a martingale.
(2) Form the process E
t
= E
Q
(B
−1
T
XF
t
).
(3) Find a previsible process φ, such that dE
t
= φ
t
dZ
t
.
Step one
The dollar discounted worth of the sterling bond is
Z
t
= C
0
exp
_
σW
t
+ (µ + u −r)t
_
.
Can we make this into a martingale under some new measure Q? Only if
˜
W
t
=
W
t
+ σ
−1
_
µ + u −r +
1
2
σ
2
_
t is a QBrownian motion, which is made possible as
before by the CameronMartinGirsanov theorem. Then, under Q
Z
t
= C
0
exp
_
σ
˜
W
t
−
1
2
σ
2
t
_
,
and thus C
t
= C
0
exp
_
σ
˜
W
t
+ (r −u −
1
2
σ
2
)t
_
.
Step two
Given this Q, deﬁne the process E
t
to be the conditional expectation process E
Q
(B
−1
T
XF
t
),
which as noted before is a Qmartingale.
Step three
The martingale representation theorem produces an Fprevisible process φ
t
linking
E
t
with Z
t
, such that
E
t
= E
0
+
_
t
0
φ
s
dZ
s
.
Now where? We need a replicating strategy (φ
t
, ψ
t
) detailing holdings of our two
dollar tradables S
t
and B
t
, so we try
• holding φ
t
units of sterling cash bond, and
• holding ψ
t
= E
t
−φ
t
Z
t
units of dollar cash bond.
The dollar value of the replicating portfolio at time t is V
t
= φ
t
S
t
+ ψ
t
B
t
= B
t
E
t
.
This portfolio is only selfﬁnancing if changes in its value are only due to changes in
the assets’ prices, that is dV
t
= φ
t
dS
t
+ ψ
t
dB
t
, or as was shown to be equivalent in
section 3.7, if dE
t
= φ
t
dZ
t
— which is precisely what the martingale representation
theorem guarantees.
80 CHAPTER 4. PRICING MARKET SECURITIES
Since V
T
= B
T
E
T
, and E
T
is the discounted claim B
−1
T
X, we have a selfﬁnancing
strategy (φ
t
, ψ
t
) which replicates our arbitrary claim X.
Option price formula (foreign exchange)
All claims have arbitrage prices and those prices are given by the portfolio value
V
t
= B
t
E
Q
(B
−1
T
XF
t
).
where Q is the measure under which the discounted asset Z
t
is a martingale.
Example — forward contract
A sterling forward contract. At what price should we agree to trade sterling at a future
date T? If we agree to buy a unit of sterling for an amount k of dollars, our payoff at
time T is
X = C
T
−k.
Its worth at time t is V
t
= B
t
E
Q
(B
−1
T
XF
t
) which is e
−r(T−t)
E
Q
(C
T
− kF
t
). So the
forward price at time zero for purchasing sterling at time T is k = E
Q
(C
T
) or
F = E
Q
_
C
0
exp
_
σ
˜
W
T
+
_
r −u −
1
2
σ
2
_
T
__
= e
(r−u)T
C
0
.
That is, the current price for sterling discounted by a factor depending on the differ
ence between the interest rates of the two currencies. With this strike, the contract’s
value at time t is
V
t
= e
−uT
_
e
ut
C
t
−e
rt
C
0
_
.
The discounted portfolio value is E
t
= B
−1
t
V
t
= e
−uT
Z
t
− e
−uT
C
0
, thus dE
t
=
e
−uT
dZ
t
, and so the required hedge φ
t
is the constant e
−uT
, and ψ is the constant
−e
−uT
C
0
.
This conﬁrms our earlier intuition.
Example — call option
A sterling call. Suppose we have a contract which allows us the option of buying a
pound at time T in the future for the price of k dollars. The dollar payoff at time T is
X = (C
T
−k)
+
.
The value of the payoff at time t is V
t
= B
t
E
Q
(B
−1
T
XF
t
). Because C
T
is lognormally
distributed we can evaluate this easily using a probabilistic result:
Lognormal call formula
If Z is a normal N(0, 1) random variable, and F, ¯ σ and k are constants, then
E
_
_
F exp
_
¯ σZ −
1
2
¯ σ
2
_
−k
_
+
_
= FΦ
_
log
F
k
+
1
2
¯ σ
2
¯ σ
_
−kΦ
_
log
F
k
−
1
2
¯ σ
2
¯ σ
_
.
4.1. FOREIGN EXCHANGE 81
As the forward price F is E
Q
(C
T
), the value of C
T
can be written in the form
F exp(¯ σX −
1
2
¯ σ
2
), where ¯ σ
2
is the variance of log C
T
, namely σ
2
T and Z is a normal
N(0, 1) under Q.
The option price at time zero is then
_
F exp
_
¯ σZ −
1
2
¯ σ
2
_
−k
_
+
, which the theorem
tells us is
V
0
= e
−rT
_
FΦ
_
log
F
k
+
1
2
σ
2
T
σ
√
T
_
−kΦ
_
log
F
k
−
1
2
σ
2
T
σ
√
T
__
.
The hedge is
φ
t
= e
−uT
Φ
_
log
F
t
k
+
1
2
σ
2
(T −t)
σ
√
T −t
_
,
ψ
t
= −ke
−rT
Φ
_
log
F
t
k
−
1
2
σ
2
(T −t)
σ
√
T −t
_
,
where F
t
is the forward sterling price at time t, F
t
= e
(r−u)(T−t)
C
t
.
The sterling investor
A sterling investor sees things differently. Were we operating in pounds we would
not be wanting dollar price processes of tradable instruments but sterling ones. The
ﬁrst of these is simply the sterling bond D
t
= e
ut
, which will be our basic unit of
account. There is also the inverse exchange rate process C
−1
t
— the worth in pounds
of one dollar. This has the value
C
−1
t
= C
−1
0
exp(−σW
t
−µt),
but it is not the sterling price of a tradable instrument, any more than C
t
was for the
dollar investor. Our other actual sterling tradable price process is the sterling value
of the dollar bond C
−1
t
B
t
.
With our two sterling tradable prices, D
t
and C
−1
t
B
t
, we can follow again our
threestep replication program, The sterling discounted value of the dollar bond is
Y
t
= D
−1
t
C
−1
t
B
t
= C
−1
0
exp(−σW
t
−(µ + u −r)t).
This discounted price process Y
t
will be a martingale under the new measure Q
£
, if
˜
W
£
t
= W
t
+ σ
−1
_
µ + u −r −
1
2
σ
2
_
t
is Q
£
Brownian motion. Then hedging will be possible as before.
Option price formula (sterling investor)
The value to the sterling investor of a sterling payoff X at time T is
U
t
= D
t
E
Q
£(D
−1
T
XF
t
).
where Q
£
is the measure under which the sterling discounted asset Y
t
is a martin
gale.
82 CHAPTER 4. PRICING MARKET SECURITIES
Change of numeraire
A worrying possibility now surfaces — the measures Q and Q
£
are different. Will
the dollar and sterling investors disagree about the price of the same security?
Suppose X is a dollar claim which pays off at time T. To the dollar investor, the
claim is worth at time t
V
t
= B
t
E
Q
_
B
−1
T
X
¸
¸
F
t
_
dollars.
To the sterling investor, the claim pays off C
−1
T
X pounds, rather than X dollars, at
time T, and its sterling worth at time t is
U
t
= D
t
E
Q
£
_
D
−1
T
(C
−1
T
X)
¸
¸
F
t
_
pounds.
Do these two prices agree? That is, is the dollar worth of the sterling valuation, C
t
U
t
,
the same as the original dollar valuation V
t
?
The Q
£
Brownian motion
˜
W
£
t
is equal to
˜
W
t
− σt, so that by the converse of
the Cameron MartinGirsanov theorem the RadonNikodym derivative of Q
£
with
respect to Q (up to time T) must be
dQ
£
dQ
= exp
_
σ
˜
W
T
−
1
2
σ
2
T
_
.
The Qmartingale associated with the RadonNikodym derivative, formed by condi
tional expectation is
ζ
t
= E
Q
_
dQ
£
dQ
¸
¸
¸
¸
F
t
_
= exp
_
σ
˜
W
t
−
1
2
σ
2
t
_
.
Note that ζ
t
is (up to a constant) the dollar discounted worth of the sterling bond.
Concretely, C
0
ζ
t
= Z
t
= B
−1
t
C
t
D
t
. Recall also (RadonNikodym fact (ii) of section
3.4) that for any random variable X which is known by time T,
E
Q
£(XF
t
) = ζ
−1
t
E
Q
(ζ
T
XF
t
).
So the dollar worth of the sterling investor’s valuation is
C
t
U
t
= C
t
D
t
E
Q
£
_
D
−1
T
C
−1
T
X
¸
¸
F
t
_
= C
t
D
t
ζ
−1
t
E
Q
_
ζ
T
D
−1
T
C
−1
T
X
¸
¸
F
t
_
,
which is (substituting in the ζ
t
expression) equal to
C
t
U
t
= B
t
E
Q
_
B
−1
T
X
¸
¸
F
t
_
= V
t
.
Thus the payoff of X dollars at time T is worth the same to either investor at any
time beforehand. Similar calculations show that the dollar and sterling investors’
replicating strategies for X are identical. So they agree not only on the prices but
also on the hedging strategy.
The difference of martingale measures only reﬂected the different numeraires of
the two investors rather than any fundamental disagreement over prices. Further
details on the effect, or lack of it, of changing numeraires are in section 6.4.
All investors, whatever their currency of account, will agree on the current value
of a derivative or other security.
4.2. EQUITIES AND DIVIDENDS 83
4.2 Equities and dividends
An equity is a stock which makes periodic cash payments to the current holder. Our
previous models treated a stock as a pure asset, but they can be modiﬁed to handle
dividend payments.
It is simplest to begin with a dividend which is paid continuously.
Equity model with continuous dividends
Let the stock price St follow a BlackScholes model, S
t
= S
0
exp(σW
t
+ ρt) and
B
t
be a constantrate cash bond B
t
= exp(rt). The dividend payment made in the
time interval of length dt starting at time t is
δS
t
dt,
where δ is a constant of proportionality.
Just as with foreign exchange, our problem is that the process S
t
is not a tradable
asset. If we buy the stock for S
0
, by the time we come to sell it at time t, what
we bought is worth not just the price of the stock itself, namely S
t
, but also the total
accumulated dividends, which under the model will depend on all the different values
that the stock has taken up until time t. The process S
t
is no longer the value of the
asset as a whole, because it is not enough.
We need to translate S
t
somehow, and to ﬁnd a new process as we did in foreign
exchange, which involves S
t
but is a tradable. Consider the following simple portfolio
strategy. The portfolio starts with one unit of stock, costing S
0
, and at every instant
when the cash dividend is paid out, that cash is immediately used to buy a little
more stock. That is, we are continuously reinvesting the dividends in the stock. The
inﬁnitesimal payout is δS
t
dt per unit of stock, which will purchase δdt more units of
stock. At time t, the number of stock units held by the portfolio will be exp(δt), and
the worth of the portfolio is
˜
S
t
= S
0
exp
_
σW
t
+ (µ + δ)t
_
.
Note how the structure of the model’s assumptions made the translation straightfor
ward. We assumed that the dividend payments were a constant proportion of the
stock price. As a consequence it made it natural to construct the tradable by reinvest
ing in the stock. If we had assumed that the dividend stream was known in advance,
independent of the stock price, then we would have reinvested in the cash bond (for
an example of this see section 4.3 on bonds). Assumptions are all.
Replicating strategies — equities
Our deﬁnition of a portfolio of stock and bond (φ
t
, ψ
t
) can be rewritten as a portfolio
of the reinvested stock and bond (
˜
φ
t
, ψ
t
), where
˜
φ
t
= e
−δt
φ
t
, with value V
t
= φ
t
S
t
+
ψ
t
B
t
=
˜
φ
t
˜
S
t
+ ψ
t
B
t
. The advantage of the new framework is that the selfﬁnancing
84 CHAPTER 4. PRICING MARKET SECURITIES
equation retains the familiar form
dV
t
=
˜
φ
t
d
˜
S
t
+ ψ
t
dB
t
,
whereas in the plain stock/bond notation, this equation would need to be modiﬁed by
the dividend cash stream, becoming dV
t
= φ
t
dS
t
+ψ
t
dB
t
+φ
t
δS
t
dt. That is, changes
in the portfolio value are due both to trading proﬁts and losses (the dS
t
and dB
t
terms)
and also to dividend payments.
Working now with our reinvested stock, as usual we want to make the discounted
asset
˜
Z
t
= B
−1
t
˜
S
t
into a martingale. Now
˜
Z
t
has SDE
d
˜
Z
t
=
˜
Z
t
_
σdW
t
+ (µ + δ +
1
2
σ
2
−r)dt
_
,
so that we want a measure Q under which
˜
W
t
= W
t
+ σ
−1
(µ + δ +
1
2
σ
2
− r)t is
Brownian motion. So under this martingale measure Q, d
˜
Z
t
− σ
˜
Z
t
d
˜
W
t
. To construct
a strategy to hedge a claim X maturing at date T, again we follow the simple Black
Scholes model, and use the martingale representation theorem. That is, there exists a
previsible process
˜
φ
t
such that
E
t
= E
Q
(B
−1
T
XF
t
) = E
Q
(B
−1
T
X) +
_
t
0
˜
φ
s
d
˜
Z
s
.
The trading strategy is to hold
˜
φ
t
units of the translated asset
˜
S
t
and ψ
t
= E
t
−
˜
φ
t
˜
Z
t
units of the cash bond. In terms of our original securities, this amounts to holding
φ
t
= e
δt
˜
φ
t
units of the stock S
t
and the same ψ
t
units of the bond B
t
.
Thus, under the martingale measure
S
t
= S
0
exp
_
σ
˜
W
t
+ (r −δ −
1
2
σ
2
)t
_
,
which is lognormally distributed.
Example — forward
An agreement to buy a unit of stock at time T for amount k has payoff
X = S
T
−k.
Its worth at time t is
V
t
= E
Q
_
e
−r(T−t)
(S
T
−k)
¸
¸
¸ F
t
_
= e
−δ(T−t)
S
t
−e
−r(T−t)
k.
The value of k which gives the contract initial nil value is the forward price of S
T
,
F = e
(r−δ)T
S
0
.
The hedge is then to hold φ
t
= e
−δ(T−t)
units of the stock and ψ
t
= −ke
−rT
units
of the bond at time t. Note the slightly surprising dynamic strategy for the forward.
Instead of simply holding a certain amount of stock until T, we are continually buying
more with the dividend income. Why? Again because of our assumption — if the
dividend payments are a known proportion of the stochastic S
t
, we have no choice
but to hide them in the stock itself.
4.2. EQUITIES AND DIVIDENDS 85
Example — call option
A call struck at k, exercised at time T has payoff X = (S
T
− k)
+
, and value at time
zero of V
0
= E
Q
_
e
−rT
(S
T
−k)
+
_
, which equals
V
0
= e
−rT
_
FΦ
_
log
F
k
+
1
2
σ
2
T
σ
√
T
_
−kΦ
_
log
F
k
−
1
2
σ
2
T
σ
√
T
__
,
where F is the forward price above e
(r−δ)T
S
0
. The hedge will be to hold e
−gd(T−t)
Φ(+)
units of the stock and have a negative holding of ke
−rT
Φ(−) units of the bond. (Here
Φ(+) and Φ(−) refer respectively to the two Φ terms in the above equation.)
Again the BlackScholes call option formula reemerges — if the martingale mea
sure Q makes the process under study, S
t
, have a lognormal distribution, then the
theorem in section 4.1 comes into play. Knowing the forward F and the term volatil
ity σ is enough to specify the price.
Example — guaranteed equity proﬁts
A contract pays off according to gains of the UK FTSE stock index S
t
, with a guar
anteed minimum payout and a maximum payout. More precisely, it is a ﬁveyear
contract which pays out 90% times the ratio of the terminal and initial values of
FTSE. Or it pays out 130% if otherwise it would be less, or 180% if otherwise it
would be more. How much is this payout worth?
Our data are
FTSE drift µ = 7%
FTSE volatility σ = 15%
FTSE dividend yield δ = 4%
UK interest rate r = 6.5%
As FTSE is composed of 100 different stocks, their separate dividend payments will
approximate a continuously paying stream. The claim X is
X = min
_
max{1.3, 0.9S
T
}, 1.8
_
,
where T is 5 years and the initial FTSE value S
0
is 1. This claim can be rewritten as
X = 1.3 + 0.9
_
(S
T
−1.444)
+
−(S
T
−2)
+
_
.
That is, X is actually the difference of two FTSE calls (plus some cash). The forward
price for S
T
is
F = e
(r−δ)T
S
0
= 1.133.
Using the above call price formula for dividendpaying stocks, we can value these
calls (per unit) at 0.0422 and 0.0067 respectively. The worth of X at time zero is then
V
0
= 1.3e
−rT
+ 0.9(0.0422 −0.0067) = 0.9712.
Were we to have forgotten that the constituent stocks of FTSE pay dividends, but
the dividends are not reﬂected in the index, we would incorrectly have valued the
contract at 1.0183 — about 5% too high.
86 CHAPTER 4. PRICING MARKET SECURITIES
Periodic dividends
In practice, an individual stock pays dividends at regular intervals rather than contin
uously, but this presents no real problems for our basic model. Let us assume that
the times of dividend payments T
1
, T
2
, . . . are known in advance, and at each time
T
i
, the current holder of the equity receives a payment of a fraction δ of the current
stock price. The stock price must also instantaneously decrease by the same amount
— or else there would be an arbitrage opportunity. At any time T = T
i
, then, we can
assume the dividend payout exactly equals the instantaneous decrease in the stock
price.
Equity model with periodic dividends
At deterministic times T
1
, T
2
, . . ., the equity pays a dividend of a fraction δ of the
stock price which was current just before the dividend is paid. The stock price
process itself is modeled as
S
t
= S
0
(1 −δ)
n[t]
exp(σW
t
+ µt),
where n[t] = max{i, T
i
≤ t} is the number of dividend payments made by time t.
There is also a cash bond B
t
= exp(rt).
We face two problems. The ﬁrst is the familiar one that S
t
is not by itself the price
of tradable asset. Translation, however, should provide a cure. The second is more se
rious. Away fromthe times T
i
, S
t
has the usual SDE of dS
t
= S
t
_
σdW
t
+ (µ +
1
2
σ
2
)dt
_
,
but at those times it has discontinuous jumps. Thus S
t
is discontinuous — it doesn’t
ﬁt our deﬁnition of a stochastic process. Fortunately, translation cures this as well.
Consider the following trading strategy. Starting with one unit of stock, every time
the stock pays a dividend we reinvest the dividend by buying more stock. At time t,
we will have (1 − δ)
−n[t]
units of the stock, and the value of our portfolio will be
˜
S
t
,
where
˜
S
t
= (1 −δ)
−n[t]
S
t
= S
0
exp(σW
t
+ µt).
As before,
˜
S
t
is tradable but our arbitrage justiﬁed assumption that the dividend pay
ments match the stock price jumps feeds through into making
˜
S
t
continuous as well.
We are back in familiar territory.
Replicating strategy
Our trading strategy will then be (
˜
φ
t
, ψ
t
), where
˜
φ
t
is the number of units of
˜
S
t
we
hold at time t, and ψ
t
is the amount of the cash bond B
t
. Such a strategy is equivalent
to holding φ
t
= (1 −δ)
−n[t]
˜
φ
t
units of the actual stock S
t
The discounted value of the (
˜
φ
t
, ψ
t
) portfolio is E
t
=
˜
φ
t
˜
Z
t
+ ψ
t
, where
˜
Z
t
is the
discounted value of the reinvested stock price
˜
Z
t
= B
−1
t
˜
S
t
. The portfolio will be
selfﬁnancing if dE
t
=
˜
φ
t
d
˜
Z
t
.
As before, we want to ﬁnd a Q which makes
˜
Z
t
into a martingale. As d
˜
Z
t
=
4.3. BONDS 87
˜
Z
t
_
σdW
t
+ (µ +
1
2
σ
2
−r)dt
_
, this will have no drift if
˜
W
t
= W
t
+ σ
−1
(µ +
1
2
σ
2
− r)t
is QBrownian motion. Then
˜
Z
t
is also a Qmartingale.
We can form the process E
t
= E
Q
(B
−1
T
XF
t
), where X is the option on the stock
which we wish to hedge.
Finally, the martingale representation theorem produces a hedging process
˜
φ
t
and
the corresponding ψ
t
can be set to be E
t
−
˜
φ
t
˜
Z
t
. So hedging is still possible in this
case, and the value at time zero of the claim X is E
Q
(B
−1
T
X).
The stock price, under Q, is
S
t
= S
0
(1 −δ)
n[t]
e
σ
˜
W
t
+(r−
1
2
σ
2
)t
.
Since this is lognormal, with the forward price for S
T
equal to F = S
0
(1 −δ)
n[T]
e
rT
,
the BlackScholes price for a call option struck at k is equal to
V
0
= e
−rT
_
FΦ
_
log
F
k
+
1
2
σ
2
T
σ
√
T
_
−kΦ
_
log
F
k
−
1
2
σ
2
T
σ
√
T
__
.
4.3 Bonds
A pure discount bond is a security which pays off one unit at some future maturity
time T. Were interest rates completely constant at rate r it would have present value
at time t of e
−r(T−t)
. We might, however, want to consider the effect of interest rates
being stochastic — much as they are in real markets. And with varying interest rates,
uncertainty about their future values would cause a discount bond price to move
randomly as well.
A full model of discount bonds, or for that matter coupon bonds, will have to
wait for chapter ﬁve and term structure models. The interplay of interest rates of
different maturities and the arbitrage mineﬁeld that models have to tiptoe through
is not something we want to worry about in a simple BlackScholes account. As
a consequence we will try to take a schizophrenic attitude to interest rates. Bond
prices will vary stochastically, but the shortterm interest rate will be deterministic.
In the real markets there is clearly a link, but then it can be argued that there are links
between stock or foreign exchange prices and the cash bonds as well. Over short
time horizons most practitioners ignore these links in all three markets.
Discount bonds
The BlackScholes model for discount bonds is:
88 CHAPTER 4. PRICING MARKET SECURITIES
Discount bond model
We assume a cash bond B
t
= exp(rt) for some positive constant r, and a discount
bond S
t
whose price follows
S
t
= S
0
exp(σW
t
+ µt),
for all times t less than T, some time horizon T long before the maturity time τ of
the bond.
In formulation, this model is indistinguishable from the simple BlackScholes
model for stocks. Thus the forward price for purchasing the bond at time T < τ
is
F = E
Q
(S
T
),
where Qis the measure under which e
−rt
S
t
is a martingale. Since σ
2
T is the variance,
under Q, of log S
T
(σ is the term volatility), then the price of a call on S
T
struck at k
is
e
−rT
_
FΦ
_
log
F
k
+
1
2
σ
2
T
σ
√
T
_
−kΦ
_
log
F
k
−
1
2
σ
2
T
σ
√
T
__
.
We have to be careful, though, with our assumption that T is much before the ma
turity τ. Not only does the distinction between the deterministic cash bond and the
stochastic discount bond get harder to maintain as T approaches τ, but for similar
reasons it gets harder to justify a simple drift µ and a constant positive σ. The bond
promises one unit at time τ, thus its price at time τ must be S
τ
= 1. In a good model,
the drift and volatility will conspire to ensure this pull to par — and indeed this will
happen in chapter ﬁve. Here if we let T = τ, we would have no such guarantee.
Bonds with coupons
Most market bonds do not just pay off one unit at maturity, but also pay off a se
ries of smaller amounts c at various predetermined times T
1
, T
2
, . . . , T
n
before matu
rity. Such coupon payments may resemble dividend payments, but unlike the equity
model, the amount of the coupon is known in advance. Here the schizophrenia ex
tends to the treatment of coupons before and after the expiry date, T, of the option.
The simplest model is to view coupons that occur before time T as coming under the
regime of the deterministic cash bond, and coupons occurring after time T (including
the redemption payment at maturity) as following a stochastic price process.
4.3. BONDS 89
Coupon bond model
There is a simple cash bond B
t
= exp(rt), and a coupon bond which pays off an
amount c at times T
1
, T
2
, . . ., up to a horizon τ. Denoting I(t) = min{i : t < T
i
}
to be the sequence number of the next coupon payment after time t, and j to be
I(T) −1, the total number of payments before time T, then the price of the bond at
time t is then
S
t
=
j
i=I(t)
ce
−r(T
i
−t)
+ Aexp(σW
t
+ µt), t < T.
Speciﬁcally, we model the ﬁrst sort of coupon (payable at, for example, T
i
< T)
to be worth
ce
−r(T
i
−t)
at time t (t < T
i
),
and for the sum of all the postT payments to evolve as an exponential Brownian
motion
Aexp(σW
t
+ µt), for t < T,
for constants A, σ, and µ.
Again S
t
is discontinuous at the coupon payment times, and again we can use a
translation rather like the one used for equity dividends (section 4.2). But because
the coupon payments are known in advance, this time we manufacture a continuous
tradable asset by holding one unit of the coupon bond and investing all the coupon
payments, as they occur, in the cash bond. The value of this asset is
˜
S
t
, where
˜
S
t
=
j
i=1
ce
−r(T
i
−t)
+ Aexp(σW
t
+ µt).
This is now a tradable asset with a continuous stochastic process.
Replicating strategy
We describe a portfolio as (
˜
φ
t
, ψ
t
), where
˜
φ
t
is the amount of the asset
˜
S
t
held at time
t, and ψ
t
is the direct holding of the cash bond B
t
= e
rt
. We let V
t
be the value of the
portfolio, V
t
=
˜
φ
t
˜
S
t
+ ψ
t
B
t
, and E
t
be its discounted value E
t
=
˜
φ
t
˜
Z
t
+ ψ
t
, where
˜
Z
t
is the discounted value of the asset
˜
S
t
. The portfolio is selfﬁnancing if dE
t
=
˜
φ
t
d
˜
Z
t
.
As usual, we want to make
˜
Z
t
into a martingale by changing measure. In fact
˜
Z
t
is just a constant cash sum of
j
i=1
ce
−rT
i
plus an exponential Brownian motion
Aexp(σW
t
+ (µ − r)t). This will be a Qmartingale if
˜
W
t
= W
t
+ σ
−1
(µ +
1
2
σ
2
− r)t
is QBrownian motion.
For an option X payable at time T, the process E
t
= E
Q
(B
−1
T
XF
t
) can be repre
sented as dE
t
=
˜
φ
t
d
˜
Z
t
for some previsible process
˜
φ
t
. We can set ψ
t
to be E
t
−
˜
φ
t
˜
Z
t
,
so that (
˜
φ
t
, ψ
t
) a hedging strategy for X. The value of X at time zero must now be
E
Q
(B
−1
T
X).
90 CHAPTER 4. PRICING MARKET SECURITIES
Under Q, the price of the bond at time T is just
S
T
= Aexp
_
σ
˜
W
T
+ (r −
1
2
σ
2
)T
_
.
This is lognormally distributed, so we can follow the call formula from section 4.1
to see that the forward price for S
T
is F = Ae
rT
and the value of a call on S
T
struck
at k is
e
−rT
_
FΦ
_
log
F
k
+
1
2
σ
2
T
σ
√
T
_
−kΦ
_
log
F
k
−
1
2
σ
2
T
σ
√
T
__
.
4.4 Market price of risk
Now is the time to tie some loose ends together. The same pattern has been repeat
ing through all the examples so far — the stochastic processes we have been using
as models in this chapter have been tied to tradable quantities only indirectly. The
foreign exchange process had to be converted from a nontradable cash process to a
tradable discount bond process. For equities, the model process had to have dividends
recombined to make it tradable. And for bonds, the coupons had to be reinvested in
the numeraire process. Underlying all this was a tradable/nontradable distinction
— we couldn’t use the martingale representation theorem to replicate claims until
we had something tradable to replicate with. But the distinction has so far been a
common sense one — can we do any better?
To some extent, yes. Some of the tradable/nontradable distinction is going to have
to be founded on goodwill. After all whether something can be traded or not in a free
market is not a mathematical decision. But if we decide on a particular process S
t
representing something truly tradable and select an appropriate discounting process
B
t
, then we can explore the market they create.
Martingales are tradables
Suppose that there is some measure Q under which the discounted tradable, Z
t
=
B
−1
t
S
t
, is a Qmartingale, what can we say about another process V
t
adapted to the
same ﬁltration F
t
such that E
t
= B
−1
t
V
t
is also a Qmartingale?
Firstly, the martingale representation theorem gives us that, as long as Z
t
has non
zero volatility, we can ﬁnd an Fprevisible process φ
t
such that
dE
t
= φ
t
dZ
t
.
Taking our cue from all the examples so far, we could create a portfolio (φ
t
, ψ
t
) where
at time t we are
• long φ
t
of the tradable S
t
,
• long ψ
t
= E
t
−φ
t
Z
t
of the tradable B
t
.
4.4. MARKET PRICE OF RISK 91
Then as before we can show that (φ
t
, ψ
t
) is a selfﬁnancing strategy, that is changes
in the value of the (φ
t
, ψ
t
) portfolio are explainable in terms of changes in value of
the tradable constituents alone. And the value of this portfolio at time t is always
exactly V
t
.
In other words we can make V
t
out of S
t
and B
t
. So it seems reasonable enough to
ennoble Vt with the title tradable as well. Being a Qmartingale after discounting is
enough to ensure that it can be made costlessly from tradables — so it might as well
be tradable itself. Of course all the derivatives that we have been constructing out of
claims have this property —E
Q
(B
−1
T
XF
t
) is always a Qmartingale.
Nonmartingales are nontradables
What about the other way round? Suppose B
−1
t
V
t
was not a Qmartingale. Then from
our deﬁnition of a martingale, there must be a positive probability at some times T
and s that E
Q
(B
−1
T
V
T
F
s
) = B
−1
s
V
s
. What would happen if V
t
were tradable and the
market stumbled into this possible ﬁltration?
Suppose we deﬁne another process U
t
by simply setting U
t
to be the cost of repli
cating the claim V
T
, that is U
t
= B
t
E
Q
(B
−1
T
V
T
F
t
). Then the terminal value of U
T
will be equal to V
T
but at time s, U
s
and V
s
, will be (possibly) different. As B
−1
t
U
t
is
a Qmartingale we can view U
t
as tradable by dint of being able to construct it from
S
t
and B
t
.
So we have two tradables, U
t
and V
t
, such that they are identical at time T but
different at some earlier time s (with positive probability). We then have an arbitrage
engine. If, say, U
s
were greater than V
s
, we could buy unlimited amounts of V and
sell unlimited amounts of U collecting the cash up front. The V − U portfolio can
be sold for nothing at time T, leaving just the (invested) cash as a guaranteed proﬁt.
And if U
s
were less than V
s
, we would run the engine in reverse.
Thus if V
t
were genuinely tradable, the market formed by S
t
, B
t
and V
t
would
contain arbitrage opportunities — something we might want to dismiss by ﬁat. To
avoid arbitrage engines, then, if B
−1
t
V
t
were not a Qmartingale, it had better not be
tradable.
We have something akin to a deﬁnition then. Within an established (complete)
market of tradable securities, there is a straightforward way of checking whether an
other process is a tradable security or not. It is tradable if its discounted price is a
martingale under the martingale measure Q, and is not tradable if it isn’t.
Tradable securities
Given a numeraire B
t
and a tradable asset S
t
, a process V
t
represents a tradable
asset if and only if its discounted value B
−1
t
V
t
is actually a Qmartingale, where Q
is the measure under which the discounted asset, B
−1
t
S
t
, is a martingale.
One way round, the process is just part of the ‘linear span’ of S
t
and B
t
; the other
way round, there is only room for two ‘independent’ tradables in a market deﬁned by
92 CHAPTER 4. PRICING MARKET SECURITIES
onedimensional Brownian motion — any more and there can be arbitrage.
xxxx; TNxxx
Exercise 4.1 If S
t
is a tradable BlackScholes stock price under the martin
gale measure Q, S
t
= exp
_
σ
˜
W
t
+ (r −
1
2
σ
2
)t
_
, with cash bond B
t
= exp(rt),
show that
(i) X
t
= S
2
t
is nontradeable,
(ii) X
t
= S
−α
t
, where α = 2r/σ
2
, is tradable.
Tradables and the market price of risk
The market price of risk is best introduced through a slight modiﬁcation of the simple
BlackScholes model. That model had stock price S
t
= S
0
exp(σW
t
+ µt), and SDE
dS
t
= S
t
_
σdW
t
+ (µ +
1
2
σ
2
)dt
_
.
We will ﬁnd it convenient, however, to deﬁne price processes by means of their SDEs,
typically
dS
t
= S
t
(σdW
t
+ µdt),
which has solution S
t
= S
0
exp
_
σW
t
+ (µ +
1
2
σ
2
)t
_
. The only difference between
these two approaches is the subtraction of
1
2
σ
2
from the drift, which can be thought of
as just a change of notation. Both forms can be equally used to deﬁne such geometric
Brownian motions, but the SDE formulation allows a greater general class of models
to be more easily considered.
Suppose then that we have a couple of tradable risky securities S
1
t
and S
2
t
, both in
the same market — that is both are functions of the same Brownian motion W
t
, and
both are deﬁned via their SDEs,
dS
i
t
= S
i
t
(σ
i
dW
t
+ µ
i
dt), i = 1, 2.
Following the discussion on tradables, we want the discounted prices of S
1
t
and S
2
t
to be martingales under the same measure Q. So assuming a simple numeraire B
t
=
exp(rt), we have that
˜
W
t
= W
t
+
_
µ
i
−r
σ
i
_
t
must be a QBrownian motion for i equal to 1 and 2. But this can only happen if the
two changes of drift are the same. That is if
µ
1
−r
σ
1
=
µ
2
−r
σ
2
.
In one of those coincidences that cause confusion, economists attach a meaning to
this quantity — if we interpret µ as the growth rate of the tradable, r as the growth
rate of the riskless bond and a as a measure of the risk of the asset, then
γ =
µ −r
σ
4.4. MARKET PRICE OF RISK 93
is the rate of extra return (above the riskfree rate) per unit of risk. As such it is often
called the market price of risk.
Using this language then gives us a simple and compelling categorization of trad
ables in terms of their SDEs — all tradables in a market should have the same market
price of risk.
The general market price of risk
We can, in fact, generalize to more sophisticated onefactor models. Rigor will have
to wait until section 6.1, but for now we can observe that a general stochastic price
process S
t
will have SDE
dS
t
= S
t
(σ
t
dW
t
+ µ
t
dt),
where σ
t
and µ
t
are previsible processes.
Then deﬁning
γ
t
=
µ
t
−r
σ
t
gives a time and state dependent market price of risk. Despite this variation, the same
as above will hold. All tradable securities must instantaneously have the same market
price of risk.
The riskneutral measure
It is worth reﬂecting on what we have done — we have provided justiﬁcation that to
be tradable in a market deﬁned by a stock S
t
and a numeraire B
t
is to share, after
discounting by B
t
, a martingale measure with S
t
. This translates naturally in SDE
terms to sharing a market price of risk — the market price of risk is actually the drift
change of the underlying Brownian motion given by CameronMartinGirsanov. So
we have a natural means for sorting through SDEs for tradables.
We also have a natural explanation for the market terminology of Q as the risk
neutral measure. If we write the SDEs in terms of the QBrownian motion
˜
W
t
:
dS
t
= S
t
_
σ
t
d
˜
W
t
+ ˜ µ
t
dt
_
,
then S
t
is tradable if and only if its market price of risk is zero. All tradables then
have the same growth rate under Q as the cash bond, independent of their riskiness
σ
t
— the measure Q is neutral with respect to risk.
But we should not overstretch the economic analogy — within our one factor
market all tradables are instantaneously perfectly correlated. They share a market
price of risk not for profound economic reasons or because investors behave with
certain risk preferences but for the reason that to do otherwise would produce a non
martingale process with a consequent opportunity for arbitrage. The market price of
risk is only a convenient algebraic form for the change of measure from P to Q, not a
new argument for using it.
94 CHAPTER 4. PRICING MARKET SECURITIES
Nontradable quantities
But convenient it is. Let’s return to our underlying theme — dealing with non
tradable processes. With foreign exchange, equities and bonds we had a model for a
process that had a ﬁxed relationship to a tradable but was itself nontradable. Con
cretely, we might have a nontradable X
t
which is modeled with the stochastic dif
ferential
dX
t
= σ
t
dW
t
+ µ
t
dt,
where σ
t
and µ
t
are previsible processes and W
t
is PBrownian motion. Here σ
t
and
µ
t
might be constants or constant multiples of X
t
, but they needn’t be.
We have X
t
nontradable but a deterministic function of X
t
, Y
t
= f(X
t
), is trad
able. Then by It ˆ o’s formula, Y has differential increment
dY
t
= σ
t
f
(X
t
)dW
t
+
_
µ
t
f
(X
t
) +
1
2
σ
2
t
f
(X
t
)
_
dt.
As Y
t
is tradable, we can write down the market price of risk for Y
t
immediately.
Assuming the discount rate is constant at r,
γ
t
=
µ
t
f
(X
t
) +
1
2
σ
2
t
f
(X
t
) −rf(X
t
)
σ
t
f
(X
t
)
Since this market price of risk is simply the change of measure from P to Q, we can
write down X’s behavior under Q as
dX
t
= σ
t
d
˜
W
t
+
rf(X
t
) −
1
2
σ
2
t
f
(X
t
)
f
(X
t
)
dt.
Thus if we have claims on X
t
, they can be priced via the normal expectation route,
using this riskneutral SDE for X
t
.
Examples
(i) If X
t
is the logarithm of a tradable asset, then f is the exponential function
f(x) = e
x
. In the simple case where σ
t
= σ and µ
t
= µ are constants (the basic
BlackScholes model), then the market price of risk for tradables is
γ
t
=
µ +
1
2
σ
2
−r
σ
,
and the corresponding riskneutral SDE for X
t
is
dX
t
= σd
˜
W
t
+
_
r −
1
2
σ
2
_
dt.
4.5. QUANTOS 95
Timedependent transforms
More generally, suppose interest rates follow the process r
t
, X
t
is nontradable with
stochastic differential
dX
t
= σ
t
dW
t
+ µ
t
dt,
and Y is a tradable security which is a deterministic function of X and time, that
is Y
t
= f(X
t
, t). Then under the martingale measure Q, X has differential
dX
t
= σ
t
d
˜
W
t
+
r
t
f(X
t
, t) −
1
2
σ
2
t
f
(X
t
, t) −∂
t
f(X
t
, t)
f
(X
t
, t)
dt,
where f
and f
are derivatives of f with respect to x, and ∂
t
f is the derivative of
f with respect to t.
(ii) The price process S
t
pays dividends at rate δS
t
. Let X
t
be the process S
t
and
assume that it follows the BlackScholes model
dX
t
= X
t
(σdW
t
+ µdt).
The asset Y
t
= exp(δt)X
t
made from instantaneously reinvesting the dividends
back into the stock holding is a tradable asset. The function f is thus chosen to
be f(x, t) = xe
δt
. The market price of risk for tradables is then
γ
t
=
µX
t
e
δt
+ δX
t
e
δt
−rX
t
e
δt
σX
t
e
δt
=
µ + δ −r
σ
,
and thus the riskneutral SDE for X
t
becomes
dX
t
= X
t
_
σd
˜
W
t
+ (r −δ)dt
_
.
(iii) Foreign exchange, the ‘wrong way round’. Let C
t
be the dollar/mark exchange
rate (worth in deutschmarks of one dollar), then the rate C
t
paid in dollars is
nontradable. (That is, if C
t
is equal to DM 1.45, the process worth $1.45 is
not tradable.) However the process 1/C
t
is tradable, or more strictly e
ut
/C
t
is a
dollar tradable asset if German interest rates are constant at rate u. If X
t
= C
t
has SDE
dX
t
= X
t
(σ
t
dW
t
+ µ
t
dt),
then the timedependent transform of f(x, t) = e
ut
/x tells us that its riskneutral
SDE is
dX
t
= X
t
_
σ
t
d
˜
W
t
+ (σ
2
t
+ u −r)dt
_
.
4.5 Quantos
British Petroleum, a UK company, has a sterling denominated stock price. But in
stead of thinking of that stock price just in pounds, we could also consider it as a pure
number which could be denominated in any currency. Contracts like this which pay
off in the ‘wrong’ currency are quantos. For instance, if the current stock price were
96 CHAPTER 4. PRICING MARKET SECURITIES
£5.20, we could have a derivative that paid this price in dollars, that is $5.20. This
is not the same as the worth of the BP stock in dollars — that would depend on the
exchange rate. What we have done is a purely formal change of units, whilst leaving
the actual number unaltered.
Quantos are best described with examples. Here are three:
• a forward contract, namely receiving the BP stock price at time T as if it were
in dollars in exchange for paying a preagreed dollar amount;
• a digital contract which pays one dollar at time T if the then BP stock price is
larger than some preagreed strike;
• an option to receive the BP stock price less a strike price, in dollars.
In each case, a simple derivative is given the added twist of paying off in a cur
rency other than in which the underlying security is denominated. And our intuition
should warn us that this act of switching currency is not a foreign exchange quibble
but something more fundamental. The British Petroleum stock price in dollars is a
meaningful concept, but it is not a traded security. The payoffs we describe involve a
nontradable quantity.
Suppose we have a simple twofactor model. We have not actually met multi
factor models yet, but they are no more problematic than singlefactor ones if we
keep our head. Rigor can be found in the multiple stock models section (6.3). Our
two random processes will be the stock price and the exchange rate, which will be
driven by two independent Brownian motions W
1
(t) and W
2
(t).
For the construction, it is helpful to recall exercise 3.2: for ρ lying between −1
and 1, then ρW
1
(t) +
_
1 −ρ
2
W
2
(t) is also a Brownian motion, and it has correlation
ρ with the original Brownian motion W
1
(t). This is a useful way to manufacture two
Brownian motions which are correlated out of a pair which are independent.
We suppose there exist the following constants: drifts µ and ν, positive volatilities
σ
1
, and σ
2
, and a correlation ρ lying between −1 and 1.
Given these constants, the quanto model is:
Quanto model
The sterling stock price S
t
and the value of one pound in dollars C
t
follow the
processes
S
t
= S
0
exp
_
σ
1
W
1
(t) + µt
_
,
C
t
= C
0
exp
_
ρσ
2
W
1
(t) + ¯ ρσ
2
W
2
(t) + νt
_
,
where ¯ ρ is the orthogonal complement of ρ, namely ¯ ρ =
_
1 −ρ
2
.
In addition there is a dollar cash bond B
t
= exp(rt) and a sterling cash bond
D
t
= exp(ut), for some positive constant interest rates r and u.
Before we tease out the tradable instruments in dollars, note the covariance of S
t
and C
t
. If we write our model in vector form, the vector randomvariable (log S
t
, log C
t
)
4.5. QUANTOS 97
is jointlynormally distributed with mean vector (log S
0
+ µ
t
, log C
0
+ ν
t
) and covari
ance matrix
_
σ
1
0
ρσ
2
¯ ρσ
2
__
t 0
0 t
__
σ
1
0
ρσ
2
¯ ρσ
2
_
=
_
σ
2
1
ρσ
1
σ
2
ρσ
1
σ
2
σ
2
2
_
t
That is, we have ensured a constant volatility for S
t
of σ
1
, a constant volatility for C
t
of σ
2
and a correlation between them of ρ.
Tradables
What are the dollar tradables? Following the intuition of the foreign exchange section
(4.1), there are three: the dollar worth of the sterling bond, C
t
D
t
; the dollar worth of
the stock, C
t
S
t
; and a dollar numeraire, the dollar cash bond B
t
.
Writing down the ﬁrst two of these tradables after discounting by the third, the
numeraire, we have Y
t
= B
−1
t
C
t
D
t
and Z
t
= B
−1
t
C
t
S
t
respectively. Their SDEs are
dY
t
= Y
t
_
ρσ
2
dW
1
(t) + ¯ ρσ
2
dW
2
(t) + (ν +
1
2
σ
2
2
+ u −r)dt
_
,
dZ
t
= Z
t
_
(σ
1
+ ρσ
2
)dW
1
(t) + ¯ ρσ
2
dW
2
(t) + (µ + ν +
1
2
σ
2
1
+ ρσ
1
σ
2
+
1
2
σ
2
2
−r)dt
_
.
(This can be checked using the nfactor It ˆ o’s formula of section 6.3.)
As in the market price of risk section, we know we want to ﬁnd a change of mea
sure to make these martingales, or equivalently a market price of risk that represents
this change of drift. As there are two sources of risk, W
1
(t) and W
2
(t), there will be
two separate prices of risk. Respectively, γ
1
(t) will be the price of W
1
(t)risk and
γ
2
(t) will be the price of W
2
(t)risk. In other words the market price of risk will be a
vector (γ
1
(t), γ
2
(t)). We want to choose these γ so that the drift terms in dY
t
and dZ
t
vanish simultaneously. Not surprisingly this means solving a pair of simultaneous
equations, or equivalently performing the matrix inversion
_
γ
1
(t)
γ
2
(t)
_
=
_
ρσ
2
¯ ρσ
2
σ
1
+ ρσ
2
¯ ρσ
2
_
−1
_
ν +
1
2
σ
2
2
+ u −r
µ + ν +
1
2
σ
2
1
+ ρσ
1
σ
2
+
1
2
σ
2
2
−r
_
.
This is a particular case of the more general result that the multidimensional market
price of risk is
γ
t
= Σ
−1
(µ −r1)
where Σ is the assets’ volatility matrix, µ is their drift vector, and 1 is the constant
vector (1, . . . , 1). More details are in section 6.3.
Here then we have a market price of risk γ
t
= (γ
1
(t), γ
2
(t)), given by
γ
1
=
µ +
1
2
σ
2
1
+ ρσ
1
σ
2
−u
σ
1
, and γ
2
=
ν +
1
2
σ
2
2
+ u −r −ρσ
2
γ
1
¯ ρσ
2
Thus under Q we can write the original processes S
t
and C
t
as
S
t
= S
0
exp
_
σ
1
˜
W
1
(t) + (u −ρσ
1
σ
2
−
1
2
σ
2
1
)t
_
,
C
t
= C
0
exp
_
ρσ
2
˜
W
1
(t) + ¯ ρσ
2
˜
W
2
(t) + (r −u −
1
2
σ
2
2
)t
_
.
98 CHAPTER 4. PRICING MARKET SECURITIES
xxxx; TNxxx
Exercise 4.2 Verify that the measure Q which has Brownian motions
˜
W
i
(t) = W
i
(t) +
_
t
0
γ
i
(s)ds (i = 1, 2) really is the martingale measure for
Y
t
and Z
t
.
Reassuringly the exchange rate process is as it was in section 4.1, given that
ρ
˜
W
1
(t) + ¯ ρ
˜
W
2
(t) is another QBrownian motion (as was proved in exercise 3.2).
But the stock price S
t
is different from what we expected. The drift has an extra
term: −ρσ
1
σ
2
. For every value of ρ (except one, namely (u − r)/σ
1
σ
2
) this stops the
dollar discounted stock price being a Qmartingale and thus prevents the price in
dollars from being tradable. And that’s precisely what our intuition warned us. There
isn’t a portfolio which is always worth a dollar amount numerically equal to the BP
stock price.
Pricing
Since we have a measure Q, under which the dollar tradables are martingales, we can
price up our quanto options.
Forward
To price the forward contract, it helps to reexpress the stock price at date T as
S
T
= exp(−ρσ
1
σ
2
T)F exp
_
σ
1
√
TZ −
1
2
σ
2
1
T
_
,
where F is the local currency forward price of S
T
, F = S
0
e
uT
, and Z is a normal
N(0, 1) random variable under Q.
Then the value of the forward at time zero in dollars is
V
0
= e
−rT
E
Q
(S
T
−k) = e
−rT
_
exp(−ρσ
1
σ
2
T)F −k
_
.
For this to be on market, that is to have a value of zero, we must set k to be F exp(−ρσ
1
σ
2
T).
This is not the same as the simple forward price F for sterling purchase. As σ
1
and σ
2
are both positive, it is clear that this quanto forward price is greater than the simple
forward price if and only if the correlation between the stock and the exchange rate
is negative.
This actually makes some sense. Suppose we assumed that the quanto forward
price was actually the same as the simple forward price F, then we could construct
the following portfolio at time zero: by going
• long C
0
exp
_
(r −u)T
_
units of the quanto forward struck at F,
• short one unit of the simple sterling forward also struck at F.
If our assumption about the quanto forward price also being F were correct then
this portfolio would be costless at time zero. At time T, this static replicating strategy
4.5. QUANTOS 99
would yield (in dollars)
C
0
exp
_
(r −u)T
_
(S
T
−F) −C
T
(S
T
−F) =
_
C
0
exp
_
(r −u)T
_
−C
T
_
(S
T
−F).
Noting that C
0
exp
_
(r − u)T
_
is the forward FX rate for C
T
, consider the effect of
negative correlation. If the stock price ends up above its forward and the FX rate is
below its forward, then the value of this portfolio is positive. And if the stock price
ends up below F and the FX rate is above its forward, then the value is also positive.
Negative correlation makes these winwin situations more likely — perfect neg
ative correlation makes them inevitable. If the quanto forward price really were F
under these circumstances it wouldn’t be hard to construct an arbitrage. For negative
ρ the quanto forward must be greater than F.
Digital
Our digital contract, I(S
T
> k) in dollars, has price V
0
= e
−rT
Q (S
T
> k), or if we
write F
Q
= F exp(−ρσ
1
σ
2
T) the quanto forward price, then
V
0
= e
−rT
Φ
_
log
F
Q
k
−
1
2
σ
2
1
T
σ
1
√
T
_
.
Again the surprise of the exp(−ρσ
1
σ
2
T) term. And in a ‘cleaner’ option. Surely the
event of S
T
being greater than k is independent of whether the option is denominated
in sterling or dollars. Indeed it is, but again replicating strategies, not expectation un
der P, price options. And replication involves the exchange rate, which is correlated
with the stock price.
Call option
Finally, we can compute the option price of e
−rT
E
Q
_
(S
T
−k)
+
_
as
V
0
= e
−rT
_
F
Q
Φ
_
log
F
Q
k
+
1
2
σ
2
1
T
σ
1
√
T
_
−kΦ
_
log
F
Q
k
−
1
2
σ
2
1
T
σ
1
√
T
__
.
Perhaps not surprisingly for a lognormal model, this is just the original Black
Scholes formula with the quanto forward.
xxxx; TNxxx
Exercise 4.3 Suppose everything remains the same, except that the stock S
t
is the price in yen of NTT, a Japanese stock, C
t
is the dollar/yen exchange
rate (the worth in yen of one dollar), and ρ is their correlation. What is the
one difference, between the sterling and yen cases, in the expression for the
quanto forward price?
Chapter 5
Interest rates
T
ime is money. A dollar today is better than a dollar tomorrow. And a dollar
tomorrow is better than a dollar next year. But just how much is that time
worth — is every day worth the same or will the price of money change from
time to time?
The interest rate market is where the price of money is set — how much does
it cost to have money tomorrow, money in a year, money in ten years? Previously
we made the modelling assumption that the cost of money is constant, but this isn’t
actually so. The price of money over a term depends not only on the length of the
term, but also on the momenttomoment random ﬂuctuations of the interest rate
market. In this way, money behaves just like a stock with a noisy price driven by a
Brownian motion.
The uncertainty of the market opens up the possibility of derivative instruments
based around the future value of money. Bonds, options on bonds, interest rate swaps,
exotic contracts on the time value of different currencies, are all derived from basic
interestrate securities, just as stock options are derived from stocks in the market.
In nominal cash terms, the market for such interestrate derivatives far outstrips that
for stock market products. Fortunately we shall still be able to calculate the prices of
these contracts on exactly the same riskfree hedging basis as before.
5.1 The interest rate market
The most basic interest rate contract is an agreement to pay some money now in
exchange for a promise of receiving a (usually) larger sum later. In general, the worth
of such a contract will depend on factors other than just the time value of money, such
as the credibility of the promisor and the perceived legality of the promise. Matters
such as creditworthiness and the like are not our concern here, and it is for the bond
market, not the interest rate market, to price them. We are solely concerned with the
time value of money for defaultfree borrowing.
This basic contract only requires two numbers to describe it — its length, or ma
turity, which records when we are to receive the later payment, and the ratio of the
size of that payment to our initial payment. We can call the maturity date on which
100
5.1. THE INTEREST RATE MARKET 101
we are paid T, and the fraction of the ﬁnal payment which is the initial, P(0, T). In
other words, one dollar at time T can be bought at time zero for P(0, T) dollars.
Discount bonds
But we can also regard the promise of a dollar as an asset, which will have some
worth at time t before T. This asset is called a discount bond, and the price P(0, T)
is its price at time zero. But it can have a different price at any other time t up to
maturity T, call it, say, P(t, T). This price P(t, T), the value at time t of receiving a
dollar at time T, is a process in time — the price process of a tradable security.
For any one maturity T, the situation is much like the stock market in that here
we have a tradable asset which has a stochastic price process. We feel we should
be able to model its behavior, and to price options on this Tbond by trading in it to
hedge them. (The only difference is the technical point that the bond evolves towards
a known value — at time T the bond is worth exactly one dollar, that is P(T, T) = 1.
Stocks don’t do things like that, but it won’t turn out to be a problem.)
But we haven’t got just one maturity. We could have written the contract for any
one of the unlimited number of possible maturity dates. This matters because the
bonds, although different, will be correlated. The tenyear bond, say, and the nine
year bond are going to move in very similar ways in the short term. Each bond cannot
just be treated in isolation as if it were a stock. This is the real challenge of the interest
rate market: the basic discount bonds are parameterized by two time indices, which
determine both the start of the contract and its end. Bond prices are thus a function
of two time variables, rather than just one, as stocks were.
The bond price graph is actually a twodimensional surface lying in threedimensional
space, which we can explore by taking twodimensional graphical sections through
it. Illustrated are sections along the lines t = 0 (ﬁgure 5.1) and T = 10 (ﬁgure 5.2).
Figure 5.1: Bond prices now Figure 5.2: Price of 10year bond vs time
Figure 5.1 is not the price process of an asset, but a graph of the current price of
a whole spectrum of different assets (the bonds of different maturities). This reﬂects
the current time value of money, quantifying exactly how much better it is to have
cash now rather than later. Generally the more distant the payment maturity date, the
less the current worth of the bond. Figure 5.2 is the price of one particular asset (the
tenyear discount bond). Now instead of a smooth graph, we have a noisy stochastic
102 CHAPTER 5. INTEREST RATES
process, up until it hits the value one at its maturity time. The start point of this graph
is the end point of ﬁgure 5.1, being the common value P(0, 10), or the worth now of
receiving a dollar in ten years’ time.
Yields
The picture in ﬁgure 5.1 is not particularly sensitive to what the market is doing.
Other than saying now is better than later, it doesn’t tell us very much on quick
inspection. A more informative measure of the market is an indication of the implied
average interest rate offered by a bond. If interest rates were constant at rate r, the
price of the Tbond at time t would be e
−r(T−t)
. In this particular case, r can be
recovered from the price P(t, T) via the formula r = −log P(t, T)/(T −t).
Interest rates are not constant, but that doesn’t stop us viewing this translation as
potentially useful. The rate we derive, R(t, T), is called the yield, and the mapping
from price to yield is onetoone for t less than T — no information is lost.
Yield
Given a discount bond price P(t, T) at time t, the yield R(t, T) is given by
R(t, T) = −
log P(t, T)
T −t
.
Thus for any given discount bond price curve, we can produce a yield curve; that
is, a graph of R(t, T) against T for some ﬁxed t.
Figure 5.3: Yield curve at t = 0 Figure 5.4: Yield curve at t = 4
While the discount bond price curve contains exactly the same information as the
yield curve, the translation is friendlier to the eye. Long dated bonds always have
lower prices, so the downwards slope of the price curve is inevitable, thus redundant.
Yield curves, on the other hand, can be increasing or decreasing functions of T,
revealing the average return of bonds stripped of the crude effects of maturity — the
term structure of the market.
The difference in yields at different maturities reﬂects market beliefs about future
interest rates. If there is a possibility that rates might be higher in the future, long
term loans will have to charge a higher rate than shortterm ones. Typically the yield
does increase with maturity, due to increased uncertainty about fardistant interest
5.1. THE INTEREST RATE MARKET 103
rates, but if current rates are high and expected to fall, the yield curve can become
‘inverted’ and long bond yields will be less than short bonds (ﬁgure 5.4). A good
model should be able to cope with both these possibilities.
Instantaneous rate
But what is the price of money now? The yield curve gives us an idea of the rate of
borrowing for each term length, but it would be convenient if we could summarize the
current cost of borrowing in a single number. What we can do is look at the current
rate for instantaneous borrowing. That is, borrowing which is paid back (nearly)
instantly. If at time t we borrow over the period from t to t + ∆t, where ∆t is a small
time increment, the rate we get is the yield R(t, t + ∆t):
R(t, t + ∆t) = −
log P(t, t + ∆t)
∆t
.
For ever smaller time increments this value more closely approximates to R(t, t),
which is the leftmost point of the yield curve at time t. We call this value the instan
taneous rate, or short rate, r
t
, which is given by both the expressions
r
t
= R(t, t),
and r
t
= −
∂
∂T
log P(t, t).
The instantaneous rate is just a process in time, free of any other parameters. Figure
5.5 shows an example short rate over ten years, corresponding to the evolution of the
10year bond in ﬁgure 5.2.
Figure 5.5: Instantaneous rate
We can sometimes see an interaction between the short rate and the bond prices if
they are correlated. In one instance, bond prices might be lower when the short rate is
higher, which can be seen in this example around the 4 year and 8 year marks, when
the short rate gets high and the bond price dips. Interestingly though, the high short
rate at t = 4 even exceeds the increased yields on longer bonds, giving an inverted
curve (ﬁgure 5.4).
The instantaneous rate is not only an important process in the interest rate mar
ket, but many models are based exclusively on its behavior, with all the other bonds
extrapolated from it.
104 CHAPTER 5. INTEREST RATES
Forwards
The shortrate process, r
t
, is not a onetoone mapping from the discount price curve
P(t, T). The translation also entails a loss of information. Just giving r
t
with no extra
prescription on how bond prices can move will not in general be enough to recover
P(t, T). Yet the instantaneous rate is convenient to work with. What we require is
a natural extension of r
t
which brings back the onetoone mapping to the prices
P(t, T) and the yields R(t, T), yet still preserves the idea of instantaneity.
Consider forward contracts, that is agreeing, at time t, to make a payment at a later
date T
1
and receive a payment in return at an even later date T
2
. We are really just
striking a forward on the T
2
bond. But what forward price should we pay?
There is a way of replicating this contract by, at time t, buying a T
2
bond and
selling a quantity, say k units, of the T
1
bond. This deal has initial cost P(t, T
2
) −
kP(t, T
1
) at time t, and will require us to make a payment of k at time T
1
, and will
give us a payment of one dollar at time T
2
. To give the contract nil initial value, we
must set k to be
k =
P(t, T
2
)
P(t, T
1
)
.
This k must be, or face arbitrage, the forward price of purchasing the T
2
bond at time
T
1
. The corresponding (forward) yield is then
−
log P(t, T
2
) −log P(t, T
1
)
T
2
−T
1
.
Were we to choose T
1
and T
2
very close together, say T
1
= T and T
2
= T + ∆t,
then as the increment ∆t became smaller this would converge to a forward rate for
instantaneous borrowing,
f(t, T) = −
∂
∂T
log P(t, T).
This rate, called simply the forward rate, is the forward price of instantaneous bor
rowing at time T. As we might expect, the ‘forward’ rate for borrowing now, at time
T = t, is exactly the current instantaneous rate, that is
f(t, t) = r
t
.
But unlike r
t
, given the forward rates f(t, T) we can recover the prices P(t, T) and the
yields R(t, T). The translation f(t, T) for our particular example is shown in ﬁgure
5.6.
Superﬁcially, the forward rate curve resembles the bond yield curve (ﬁgure 5.3).
Indeed the yield curve and the forward curve agree at their leftmost point, the instan
taneous rate, but the other points of the two curves will generally be different. But
the formula for R(t, T) can be differentiated and rearranged to show that
f(t, T) = R(t, T) + (T −t)
∂R
∂T
(t, T).
This tells us that the forward rate curve is higher than the yield curve, if the yield
curve is increasing, and lower than it if the yield curve is inverted.
5.2. A SIMPLE MODEL 105
Figure 5.6: Forward rate curve at time t = 0
As a function of time rather than maturity the forward rate will not be so smooth,
but will start with some initial value f(0, T) and evolve as a stochastic process, ﬁn
ishing with the value r
T
at time T.
Summary
We have a market of defaultfree zero coupon discount bonds. The price at time t
of the Tbond which pays off one dollar at time T is P(t, T). The average yield of
the bond over its remaining lifetime is R(t, T), and the price now of instantaneous
borrowing at time T is the forward rate f(t, T). The price of instantaneous spot
borrowing is r
t
= R(t, t) = f(t, t).
Both of these associated families of rates, R(t, T) and f(t, T), contain all the orig
inal price information which can be recovered. Explicitly
'
&
$
%
Interestrate market summary
The forward rates and the yield can be written in terms of the bond prices as
f(t, T) = −
∂
∂T
log P(t, T), and R(t, T) = −
log P(t, T)
T −t
.
And conversely, the bond prices can be given in terms of the forward rates or the
yields:
P(t, T) = exp
_
−
_
T
t
f(t, u)du
_
,
and P(t, T) = exp
_
−(T −t)R(t, T)
_
.
In other words, for modeling purposes we can choose to specify the behavior of
only any one of these three, and the other two will follow automatically.
5.2 A simple model
A concrete example is illuminating. The secret of this chapter is that we can tackle
interest rate models in exactly the same way as stock models. The It ˆ o manipula
106 CHAPTER 5. INTEREST RATES
tions are harder but they are not signiﬁcant for the story — just as in BlackScholes,
the real work is carried by the martingale representation theorem. In BlackScholes,
there were only two canonical tradables (the stock and the bond), but there are now
an inﬁnite number of underlying discount bonds. To pick just two of these tradables
to work with would seem to favor that pair over the rest, but such worries will prove
illusory. All the tradables will still turn out to be martingales under the riskneutral
measure, which itself is independent of the apparent ‘choice’ of instruments to work
with.
Simple interest rate model
Given an initial Tintegrable forward rate curve f(0, T), the forward curve evolves
as:
d
t
f(t, T) = σdW
t
+ α(t, T)dt,
for some constant volatility σ and drift α, a bounded deterministic function of time
and maturity.
We have set the market up not with an SDE for the price of any asset, but with the
SDE for the forward rate. However as chapter four has shown, as long as we are an
It ˆ o step away from the price of something, this doesn’t have to pose a problem.
The forward rate itself is
f(t, T) = σW
t
+ f(0, T) +
_
t
0
α(s, T)ds.
Thus the forward rate is normally distributed. Moreover the forward rates at different
maturities are perfectly correlated in their movements as the difference between any
two of them, f(t, T) − f(t, S), is purely deterministic. There is only one source
of randomness, the Brownian motion W
t
, and that is a process over time, not over
maturity.
Tradable securities
Tradables may only be an It ˆ o step away, but what are they? One is obvious — we
want a numeraire. Though chapter six will show the choice of numeraire doesn’t
really matter, there is a canonical candidate — the cash bond formed by the instanta
neous rate r
t
. That is, B
t
given by
dB
t
= r
t
B
t
dt, B
0
= 1.
Since r
t
is given by f(t, t), we can write down its integral equation easily enough as
r
t
= σW
t
+ f(0, t) +
_
t
0
α(s, t)ds.
[Technical trap: the SDE for r
t
is not just the SDE for f(t, T) ‘evaluated’ at T = t. It ˆ o
tells us it is actually dr
t
= d
t
f(t, t) +
∂f
∂T
(t, t).] Unlike the basic stock models, this rate
is not constant but rather is a random process, and it is normally distributed, which
5.2. A SIMPLE MODEL 107
admits the possibility of it being negative sometimes. Later we will show models
which overcome this, but for the moment we’ll pay this price for simplicity. Now for
the cash bond, B
t
= exp
_
_
t
0
r
s
ds
_
, which has the slightly daunting expression
B
t
= exp
_
σ
_
t
0
W
s
ds +
_
t
0
f(0, u)du +
_
t
0
_
t
s
α(s, u)duds
_
.
This will be our tradable numeraire.
What about another tradable? Here, as mentioned earlier, there’s an embarrass
ment of tradables, but let’s pick one. Fixing T, we have the price of the Tmaturity
bond P(t, T) given by P(t, T) = exp
_
−
_
T
t
f(t, u)du
_
or
P(t, T) = exp −
_
σ(T −t)W
t
+
_
T
t
f(0, u)du +
_
t
0
_
T
t
α(s, u)duds
_
.
Replicating strategies
Suppose then we wanted to replicate some claim X at a time horizon S less than T
(so that the Tmaturity bond doesn’t vanish on us). In chapters two, three and four
we had a threestage replicating strategy, so at least a ﬁrst guess would be to follow
it here as well:
'
&
$
%
Three steps to replication (interestrate market)
(1) Find a measure Q under which the Tbond discounted by the cash bond Z
t
=
B
−1
t
P(t, T) is a martingale.
(2) Form the process E
t
= E
Q
_
B
−1
S
XF
t
_
.
(3) Find a previsible process φ
t
, such that dE
t
= φ
t
dZ
t
.
First we tackle the complicatedlooking discounted bond price process Z
t
= B
−1
t
P(t, T):
Z
t
= exp −
_
σ(T −t)W
t
+ σ
_
t
0
W
s
ds +
_
T
0
f(0, u)du +
_
t
0
_
T
s
α(s, u)duds
_
.
Noting that the σ(T − t)W
t
term’s differential can be handled with the product rule
and everything else inside the exponential is either constant or easy to differentiate,
It ˆ o gives us the SDE for Z
t
as
dZ
t
= Z
t
_
−σ(T −t)dW
t
−
_
_
T
t
α(t, u)du
_
dt +
1
2
σ
2
(T −t)
2
dt
_
.
Now we are on familiar ground. Though we are used to the cash bond B
t
being
deterministic, this (random) B
t
and the Tbond price P(t, T) are both adapted to the
same Brownian motion, W
t
, and ﬁnding Z
t
doesn’t pose any real problem.
108 CHAPTER 5. INTEREST RATES
Step one
We have an SDE for Z
t
and we want to see if we can ﬁnd a change of measure drift
γ
t
for the Brownian motion which makes Z
t
driftless. The candidate is clearly
γ
t
= −
1
2
σ(T −t) +
1
σ(T −t)
_
T
t
α(t, u)du.
And since it is bounded up to time S, the technical conditions of CMG are satisﬁed
— our candidate passes and we have a measure Q, equivalent to P, such that
˜
W
t
=
W
t
+
_
t
0
γ
s
ds is a QBrownian motion. The SDE for the discounted price Z
t
now
becomes
dZ
t
= −σZ
t
(T −t)d
˜
W
t
.
The process Z
t
has no drift, and because σ(T − t) is bounded up to time S, Z
t
is a
Qmartingale.
Step two
This gives us E
t
as the conditional Qexpectation of the discounted claim B
−1
S
X,
namely
E
t
= E
Q
(B
−1
S
XF
t
)
But since E
t
is a Qmartingale just as Z
t
is, we take:
Step three
Using the martingale representation theorem to link them via an Fprevisible process
φ
t
:
E
t
= E
Q
(B
−1
S
X) +
_
t
0
φ
s
dZ
s
.
What is our trading strategy? At time t,
• hold φ
t
units of the Tbond P(t, T)
• hold ψ
t
= E
t
−φ
t
Z
t
units of the cash bond B
t
.
The undiscounted value of this portfolio at time t is
V
t
= φ
t
P(t, T) + ψ
t
B
t
= B
t
E
t
.
As before, it is also true that dV
t
= φ
t
d
t
P(t, T) + ψ
t
dB
t
and thus this portfolio is self
ﬁnancing. The strategy has an initial cost of V
0
= E
Q
(B
−1
S
X) and has a terminal
value V
S
= X, which exactly hedges the claim. Arbitrage has won through.
Option price formula (interest rate)
The price of X at time t is
V
t
= B
t
E
Q
(B
−1
S
XF
t
).
5.2. A SIMPLE MODEL 109
No free lunches
So far, so good — even though the It ˆ o work was harder, we have just another stock
type model. The chosen pair B
t
and P(t, T) behaved like any of the tradables of
chapter four. But something should worry us. We picked a particular bond, the
Tmaturity bond, and found a change of measure particular to that. Yet all claims
which paid off at time S before T could be hedged, even those, for example, which
are identical to bond of other maturities.
So we have two ways of pricing the Sbond at time t, P(t, S). One direct from its
SDE. And the other indirect, viewing X = P(S, S) = 1 as a claim to be hedged via
the cash bond and the Tbond.
There is no obvious reason why they should be the same given our original model.
And yet the same they must be. If the hedge price were ever, say, less than P(t, S)
we would have an arbitrage engine capable of locking in unlimited proﬁts. We don’t
want free lunches, so we should assume that the real world forbids them. That is we
should impose on our real world model some suitable condition to make the various
ways of getting at the price P(t, S) agree. What condition?
Consider the discounted process of the Sbond, Y
t
= B
−1
t
P(t, S). Reworking the
It ˆ o from before we have, as expected,
dY
t
= Y
t
_
−σ(S −t)dW
t
−
_
_
S
t
α(t, u)du
_
dt +
1
2
σ
2
(S −t)
2
dt
_
.
If we deﬁne γ
S
t
to be
γ
S
t
= −
1
2
σ(S −t) +
1
σ(S −t)
_
S
t
α(t, u)du,
then we have dY
t
= −σY
t
(S −t)(dW
t
+γ
S
t
dt), or in terms of the QBrownian motion
we had before:
dY
t
= −σY
t
(S −t)
_
d
˜
W
t
+ (γ
S
t
−γ
t
)dt
_
.
This discounted process must be a Qmartingale — it’s tradable and, from the risk
free hedging construction, Y
t
= B
−1
t
P(t, S) = E
Q
(B
−1
S
F
t
). So the drift term of the
SDE above must be zero: γ
S
t
= γ
t
.
Here is the restriction we require — our arbitrary choice of T must not have af
fected the process γ
t
. So γ
t
must be independent of T , or in other words
∂γ
∂T
= 0.
Multiplying the formula for γ
t
by σ(T − t) and differentiating with respect to T,
we get:
Restriction on the drift
In an arbitragefree market, the drift α(t, T) satisﬁes
α(t, T) = σ
2
(T −t) + σγ
t
.
110 CHAPTER 5. INTEREST RATES
This equation is saying something we did not encounter, at this level, in the stock
market. It says that there are restrictions on the drifts which the forward rates can
have if there is to be no arbitrage. The drift α(t, T) may have started off as a general
deterministic function of both time and maturity, but now it is expressed as the sum
of a particular function (σ
2
(T −t)) and a process which has no maturity dependence
at all (σγ
t
). Most general functions cannot be written in this way.
In another sense, this is actually familiar ground. We can think of the SDE for
P(t, T) under P as
d
t
P(t, T) = P(t, T)
_
−σ(T −t)dW
t
+
_
r
t
−σ(T −t)γ
t
_
dt
_
.
Written this way, γ
t
stands revealed as the market price of risk (see section 4.4). We
know that every security in the market has to have the same market price of risk,
which explains why γ
t
does not depend on the maturity T chosen.
Two things stand out. Firstly there is a measure Q which makes a martingale not
just out of one discounted bond, but each and every discounted bond simultaneously.
We worried about the embarrassment of bonds to choose, but we needn’t have. There
was only one Brownian motion and that is what matters. If we freeze time and look
at just one t, the values of the bonds P(t, T) are just deterministic transformations of
each other. And if one bond can be brought into line by a change of measure then so
can they all.
If, that is, they are roughly in step in the ﬁrst place. Our second point is that there
is a price to pay for this success. If we write the original SDE for f(t, T) in terms of
the QBrownian motion,
˜
W
t
, we have:
d
t
f(t, T) = σd
˜
W
t
+ σ
2
(T −t)dt.
As we expect from a BlackScholes upbringing, the drift α(t, T) has vanished. But
α(t, T) must be recoverable by a change of measure γ
t
which has no dependence on
T. So we weren’t free to pick α(t, T) as any function of t and T — we must, unlike
BlackScholes, have some structure to the original real world drift.
But even if our success has brought slight complications, we have nonetheless
succeeded. We have a model with stochastic interest rates which is still arbitrage
complete. All claims can be coherently hedged by the underlying bonds. Once more,
replication provides the price.
5.3. SINGLEFACTOR HJM 111
'
&
$
%
Bonds and rates in terms of the QBrownian motion
˜
W
t
The bond prices, forward and short rates are given by:
P(t, T) = exp −
_
σ(T −t)
˜
W
t
+
_
T
t
f(0, u)du +
1
2
σ
2
T(T −t)t
_
,
B
t
= exp
_
σ
_
t
0
˜
W
s
ds +
_
t
0
f(0, u)du +
1
6
σ
2
t
3
_
,
f(t, T) = σ
˜
W
t
+ f(0, T) + σ
2
_
T −
1
2
t
_
t,
r
t
= σ
˜
W
t
+ f(0, t) +
1
2
σ
2
t
2
.
5.3 Singlefactor HJM
From the particular to the general. We know the basic idea — all three descriptions of
the yield curve, the prices P(t, T), the yields R(t, T) and the forward rates f(t, T) are
equivalent, so we select one and specify its behavior. HeathJarrowMorton (HJM) is
a powerful, technically rigorous interestrate model based on the instantaneous for
ward rates f(t, T).
Singlefactor HJM model
Given an initial forward rate curve f(0, T), the forward rate for each maturity T
evolves as
f(t, T) = f(0, T) +
_
t
0
σ(s, T)dW
s
+
_
t
0
α(s, T)ds, 0 ≤ t ≤ T,
or in differential form
d
t
f(t, T) = σ(t, T)dW
t
+ α(t, T)dt.
The volatilities σ(t, T) and the drifts α(t, T) can depend on the history of the Brow
nian motion W
t
and on the rates themselves up to time t.
For any ﬁxed maturity T, the forward rate evolves according to its own volatility
σ(t, T) and its own drift α(t, T). In section 5.5 we will allow the decoupling that
comes when rates can move with less than perfect correlation, but here only a single
process, a PBrownian motion W
t
, will drive each and every rate. The incremental
changes of all forward rates, and thus all yields and all bond prices are perfectly
correlated.
Our formal description is vague about the precise properties of the volatility and
drift functions. The general HJM model posits very few overarching conditions on
the σ and α, but imposes piecemeal technical constraints from time to time. Col
lected, and simpliﬁed somewhat, these technical conditions are shown in the box.
The ﬁrst two conditions make sure that the forward rates f(t, T) are well deﬁned
112 CHAPTER 5. INTEREST RATES
by their SDE. The last two conditions will be used for a Fubinitype result that the
stochastic differential of the integral of f(t, T) with respect to T is the integral of the
stochastic differentials of f. Given these box conditions, the ﬁrst three conditions of
the HJM model (C1–C3 in their paper) are satisﬁed.
Singlefactor HJM: conditions on the volatility and drift
We assume that
• for each T, the processes σ(t, T) and α(t, T) are previsible and depend only
on the history of the Brownian motion up to time t, and are good integrators
in the sense that
_
T
0
σ
2
(t, T)dt and
_
T
0
α(t, T)dt are ﬁnite;
• the initial forward curve, f(0, T), is deterministic and satisﬁes the condition
that
_
T
0
f(0, u)du < ∞;
• the drift α has ﬁnite integral
_
T
0
_
u
0
α(t, u)dtdu;
• the volatility σ has ﬁnite expectation E
_
T
0
¸
¸
_
u
0
σ(t, u)dW
t
¸
¸
du.
Numeraire
As chapter six will show, the choice of numeraire is arbitrary — but algebraic con
venience certainly points to a canonical choice. Our description of the forward
rates f(t, T) allows us to write down an integral equation for the instantaneous rate
r
t
= f(t, t) (which need not be Markov), namely:
r
t
= f(0, t) +
_
t
0
σ(s, t)dW
s
+
_
t
0
α(s, t)ds.
The simplest cash product is then the account, or bond, formed by starting with one
dollar at time zero and reinvesting continually at this rate. In other words, the bond
B is a stochastic process satisfying the SDE
dB
t
= r
t
B
t
dt, B
0
= 1, or B
t
= exp
__
t
0
r
s
ds
_
.
Integration then gives us
B
t
= exp
__
t
0
__
t
s
σ(s, u)du
_
dW
s
+
_
t
0
f(0, u)du +
_
t
0
_
t
s
α(s, u)duds
_
.
Here we used the last technical condition of the HJM box to say that the integrals
of
_
t
0
_
_
t
s
σ(s, u)dW
s
_
du can be interchanged to
_
t
0
_
_
t
s
σ(s, u)du
_
dW
s
. We have a
numeraire.
Bond prices
We need tradable assets — and we have them, the bonds P(t, T). Since the for
ward rates f(t, T) are a onetoone transformation, the bond prices themselves are
5.3. SINGLEFACTOR HJM 113
contained in the forward rate information as
P(t, T) = exp
_
−
_
T
t
f(t, u)du
_
,
which will be continuous in t and T. If we integrate the original equation for the
forward rates f(t, T) then we have the bond price P(t, T) equal to
exp −
_
_
t
0
_
_
T
t
σ(s, u)du
_
dW
s
+
_
T
t
f(0, u)du +
_
t
0
_
T
t
α(s, u)duds
_
.
Reassuringly this expression, although awkward, has the right values at time zero
(namely exp
_
−
_
T
0
f(0, u)du
_
) and time T (namely one).
Discounted bonds
Let’s ﬁx one particular maturity T to work with for the moment. As everywhere else,
our attention focuses on the discounted asset price — that is, Z(t, T) = B
−1
t
P(t, T).
By combining the above expressions for the cash bond and the bond price itself, we
get
Z(t, T) = exp
_
_
t
0
Σ(s, T)dW
s
−
_
T
0
f(0, u)du −
_
t
0
_
T
s
α(s, u)duds
_
,
where Σ(t, T) is just notation for the integral −
_
T
t
σ(t, u)du. It ˆ o handleturning then
gives the SDE —
d
t
Z(t, T) = Z(t, T)
_
Σ(t, T)dW
t
+
_
1
2
Σ
2
(t, T) −
_
T
t
α(t, u)du
_
dt
_
,
revealing the variable Σ(t, T) to be the logvolatility of P(t, T).
Change of measure
In the usual way, we want to make the discounted bond into a martingale by changing
measure. The change of measure drift (market price of risk) is
γ
t
=
1
2
Σ(t, T) −
1
Σ(t, T)
_
T
t
α(t, u)du.
We need the technical CameronMartinGirsanov theoremcondition that E
P
exp
1
2
_
T
0
γ
2
t
dt
is ﬁnite. Then there will be a new measure Q equivalent to P, such that
˜
W
t
=
W
t
+
_
t
0
γ
s
ds is QBrownian motion. The SDE of the discounted bond under Q is
then
d
t
Z(t, T) = Z(t, T)Σ(t, T)d
˜
W
t
,
which is driftless. For this to be a proper Qmartingale, it is sufﬁcient that the expo
nential martingale condition E
Q
exp
1
2
_
T
0
Σ
2
(t, T)dt < ∞holds (see section 3.5).
114 CHAPTER 5. INTEREST RATES
Bond price SDE
Under this martingale measure, the bond price P nowhas the stochastic differential
d
t
P(t, T) = P(t, T)
_
Σ(t, T)d
˜
W
t
+ r
t
dt
_
.
The concrete model of section 5.2 partially spoilt the surprise, but we have our
BlackScholes like result, even here with a general interestrate model such as HJM.
The behavior of the price P under the martingale measure does not depend on the
drift α, but only on the volatility Σ (itself a function of σ). Just as the BlackScholes
stock model under Q had no dependence on the original stock drift µ.
Replicating strategies
We’ve jumped slightly ahead of ourselves, we have found the martingale measure
and the process for P(t, T) under it. But we ought to check that we can produce
replicating strategies for claims. Suppose we have a claim X which pays off at time
S. If we are going to hedge this with a discount bond maturing at date T, our only
restriction is that S should come before T — we cannot hedge a longterm prod
uct with a shorterterm instrument. (Unless we split the timeperiod up into shorter
subsections, and roll over shortterm bonds from section to section.) Suppose, for
simplicity, we choose to use a bond with maturity T larger than S.
As before, our second step to replication is to form the conditional Qexpectation
of the discounted claim B
−1
S
X, rather than the raw claim X. That is, we deﬁne E
t
to
be the Qmartingale
E
t
= E
Q
_
B
−1
S
XF
t
_
.
For the martingale representation theorem to be used, we also need that the bond
volatility Σ(t, T) is never zero before T, in which case, we apply the representation
theorem to the martingale Z(t, T) and the discounted claim process E
t
. This gives us
that
E
t
= E
0
+
_
t
0
φ
s
dZ(s, T),
for some Fprevisible process φ.
Our trading strategy will be a combination of both a holding in the Tbond and a
holding in the cash bond B
t
. Speciﬁcally, we
• hold φ
t
units of the Tbond at time t,
• hold ψ
t
:= E
t
−φ
t
Z(t, T) units of the cash bond at time t.
The value of this portfolio at time t is
V
t
= B
t
E
t
= B
t
E
Q
_
B
−1
S
XF
t
_
.
5.3. SINGLEFACTOR HJM 115
The strategy (φ
t
, ψ
t
) will be selfﬁnancing if dV
t
= φ
t
d
t
P(t, T) + ψ
t
dB
t
, or equiva
lently (as in section 3.7) if
dE
t
= φ
t
d
t
Z(t, T).
Which is ensured by the representation of E
t
in terms of φ
t
. The portfolio (φ
t
, ψ
t
) is
selfﬁnancing. Thus:
'
&
$
%
Derivative pricing
If X is the payoff of a derivative maturing at time T, then its value at time t is
V
t
= B
t
E
Q
_
B
−1
T
XF
t
_
= E
Q
_
exp
_
−
_
T
t
r
s
ds
_
X
¸
¸
¸
¸
¸
F
t
_
.
Arbitragefree market
But the Sbond is simply a claim of X = 1 maturing at time S. Thus its worth at time
t must be B
t
E
Q
(B
−1
S
F
t
). Or more fully,
P(t, S) = E
Q
_
exp
_
−
_
S
r
r
u
du
_¸
¸
¸
¸
¸
F
t
_
, t ≤ S < T.
The martingale measure brings a pleasant simplicity. All bond prices are just the
expectation under Q of the instantaneous rate discount from t to their maturity.
What about the discounted Sbond, Z(t, S) = B
−1
t
P(t, S)? This can now be writ
ten as
Z(t, S) = E
Q
_
B
−1
S
F
t
_
.
Just as we saw in the simple example (section 5.2), all the other (discounted) bonds
are now martingales under the same Q. Which means that their drifts under P are
restricted by the need to be a simple change of measure away from a martingale. In
other words, the market price of risk has to be the same for all bonds, or else there
will be an arbitrage opportunity.
So we have a restriction on the bonds’ Pdrifts. In particular, it must be the case
that, for all maturities T,
_
T
t
α(t, u)du =
1
2
Σ
2
(t, T) −Σ(t, T)γ
t
, t ≤ T.
Differentiating with respect to T, we see that α(t, T) = −σ(t, T)Σ(t, T) + σ(t, T)γ
t
,
that is
α(t, T) = σ(t, T)
_
γ
t
−Σ(t, T)
_
.
Exactly as in section 5.2, where σ(t, T) = σ and Σ(t, T) = −σ(T − t), the real world
drift α(t, T) cannot be too different from the riskneutral value of −σ(t, T)Σ(t, T).
Under this riskneutral measure, the forward rate and the instantaneous rate are
then,
116 CHAPTER 5. INTEREST RATES
Forward and short rates under Q
d
t
f(t, T) = σ(t, T)d
˜
W
t
−σ(t, T)Σ(t, T)dt,
r
t
= f(0, t) +
_
t
0
σ(s, t)d
˜
W
s
−
_
t
0
σ(s, t)Σ(s, t)ds.
Like the bond price itself, these expressions no longer depend on the drift at all,
but are solely expressed in terms of the volatilities σ and Σ.
Model conditions
We have been accumulating technical conditions as we have swept through. They are
summarized in the box below.
The ﬁrst condition is actually necessary and sufﬁcient for there to be an equiva
lent measure under which every single discounted bond price is a martingale, which
guarantees the absence of arbitrage. The second condition is equivalent to asserting
that the change of measure is unique, which means that all risks can be hedged using
the martingale representation theorem. The last two conditions are technical require
ments for CMG to operate and to make sure that Z is a martingale under the new
measure.
Singlefactor HJM: market completeness conditions
It is required that
• there exists an Fprevisible process γ
t
, such that
α(t, T) = σ(t, T)
_
γ
t
−Σ(t, T)
_
, for all t ≤ T;
• the process A
t
= Σ(t, T) is nonzero for almost all (t, ω), t < T, for every
maturity T;
• the expectation Eexp
1
2
_
T
0
γ
2
t
dt is ﬁnite;
• and the expectation exp
1
2
_
T
0
_
γ
t
−Σ(t, T)
_
2
dt is ﬁnite.
The importance of the ﬁrst condition in this box is the constraint it places on the
drift α(t, T). As the process γ
t
is only a function of time and not of maturity, the drift
is forced to take the value −σ(t, T)Σ(t, T) modiﬁed only by the ‘onedimensional’
displacement γ
t
σ(t, T). Given that σ(t, T) and Σ(t, T) are determined by the for
ward rate volatilities, the only degree of freedom for the drift comes from the one
parameter γ
t
process. Unlike simple asset models, not all drift functions α(t, T) are
allowable.
5.4. SHORTRATE MODELS 117
5.4 Shortrate models
Shortrate models are popular in the market. In particular, they are often used to
price derivatives which depend only on one underlying bond. They have evolved
from various historical starting points — some emerging from discrete frameworks,
others from equilibrium models — and are often presented in a simple hierarchy with
no apparent connection to any overarching model.
All however are HJM models, which is why we used this framework in the ﬁrst
place. And there is a mathematical transformation that makes these two alternative
descriptions equivalent. Demonstrating that is the purpose of this section.
A shortrate model posits a riskneutral measure Qand a shortrate process r
t
. The
model is that instantaneous borrowing can take place at rate r
t
for an inﬁnitesimal
period. Rolling up the periods gives rise to a cash bond process B
t
= exp
_
_
t
0
r
s
ds
_
,
as in the HJM model. As with the equations at the end of section 5.3, the bond prices
are given by
P(t, T) = E
Q
_
exp
_
−
_
T
t
r
s
ds
_¸
¸
¸
¸
¸
F
t
_
,
and the value at time t of a claim X maturing at date T is
V (t, T) = E
Q
_
exp
_
−
_
T
t
r
s
ds
_
X
¸
¸
¸
¸
¸
F
t
_
.
The paradigm of shortrate modeling is to work within a parameterized family of
processes, which typically are Markovian. The parameters are chosen to best ﬁt the
market, and then the above expression for V
t
is calculated to price the claim X.
HJM in terms of the short rate
It is not immediately clear that this is an HJM model. To prove this requires choosing
the forward volatility surface σ(t, T) so that the resulting short rate from the HJM
model is exactly the same as the original process r
t
. This is possible for any general
short rate r
t
, but it’s easiest to show in the special case where r
t
is a Markov process.
Suppose that that r
t
is a Markov diffusion (though not necessarily timehomogeneous)
with volatility ρ(r
t
, t) and drift ν(r
t
, t). That is
dr
t
= ρ(r
t
, t)dW
t
+ ν(r
t
, t)dt,
where ρ(x, t) and ν(x, t) are deterministic functions of space and time.
Then
_
T
t
f(t, u)du = −log P(t, T) = g(r
t
, t, T) where g(x, t, T) is the deterministic
function
g(x, t, T) = −log E
Q
_
exp
_
−
_
T
t
r
s
ds
_¸
¸
¸
¸
¸
r
t
= x
_
.
There is a theorem:
118 CHAPTER 5. INTEREST RATES
Shortrate model in HJM terms
The required volatility structure is
σ(t, T) = ρ(r
t
, t)
∂
2
g
∂x∂T
(r
t
, t, T),
and Σ(t, T) = −ρ(r
t
, t)
∂g
∂x
(r
t
, t, T).
We can see why this is so, by thinking of the forward rate f(t, T) as
∂g
∂T
(r
t
, t, T),
and using It ˆ o to deduce that
d
t
f(t, T) =
∂
2
g
∂x∂T
_
ρ(r
t
, t)dW
t
+ ν(r
t
, t)dt
_
+
∂
2
g
∂t∂T
dt +
1
2
∂
3
g
∂
2
x∂T
ρ
2
(r
t
, t)dt.
The volatility term must match σ(t, T), which gives us the result. In addition, the
initial forward rate curve f(0, T) is given by
f(0, T) =
∂g
∂T
(r
0
, 0, T).
This volatility structure and initial curve then identiﬁes an HJM model for this market
with the same short rate under Q.
Short rate in terms of HJM
Conversely, it is also true that HJM models are shortrate models. The equation for
the bond price in terms of r
t
holds (see near the end of section 5.3), with r
t
in terms
of the HJM volatilities σ(t, T) and Σ(t, T), given as
r
t
= f(0, t) +
_
t
0
σ(s, t)dW
s
−
_
t
0
σ(s, t)Σ(s, t)ds.
This formula is not necessarily simple.
Ho and Lee
Now for the accepted hierarchy of shortrate models: starting with Ho and Lee. In its
shortrate form, Ho and Lee gives the SDE for r
t
under Q, the martingale measure, as
Ho and Lee model
The short rate is driven by the SDE:
dr
t
= σdW
t
+ θ
t
dt,
for some θ
t
deterministic and bounded, and σ constant.
The question we should immediately ask is, which HJM model corresponds to
this? Following the mechanics from earlier, we ﬁnd via It ˆ o
g(x, t, T) = x(T −t) −
1
6
σ
2
(T −t)
3
+
_
T
t
(T −s)θ
s
ds.
5.4. SHORTRATE MODELS 119
Thus σ(t, T), the HJM volatility surface, is simply σ
∂
2
g
∂x∂T
= σ. Thus the volatility
surface is constant, depending on neither time nor maturity. We can fully specie the
HJM model under Q as
Ho and Lee model in HJM terms
d
t
f(t, T) = σdW
t
+ σ
2
(T −t)dt
with f(0, T) =
∂g
∂T
(r
0
, 0, T) = r
0
−
1
2
σ
2
T
2
+
_
T
0
θ
s
ds.
Equivalently, we can provide the evolution of the bond prices P(t, T) under Q:
P(t, T) = exp −
_
σ(T −t)W
t
+
_
T
t
f(0, u)du +
1
2
σ
2
T(T −t)t
_
.
This model is the (general) singlefactor model with constant volatility, and is actu
ally the simple model of section 5.2. If used in the shortrate form, then σ sets the
volatility of all forward rates and θ
s
allows matching to any initial forward curve via
the identity f(0, T) = r
0
−
1
2
σ
2
T
2
+
_
T
0
θ
s
ds.
It is a simple model, and its simplicity tells against it — the forward rates and the
short rate r
t
can go negative occasionally, and go to inﬁnity in the long term. And
not just of course under Q, but under any equivalent measure P as well. Many other
models expend much effort just to avoid these pitfalls.
But it is not that simple a model — the HJM formulation allows a description of
how the real forward curve can move over time. Given any previsible process γ
t
, the
forward rates can move as
d
t
f(t, T) = σdW
t
+
_
σ
2
(T −t) + σγ
t
_
dt,
with dr
t
= σdW
t
+ (θ
t
+ σγ
t
)dt.
So short rates can have a wide range of possible drifts under P the real world measure,
not just the simple deterministic drift θ
t
. The restriction with Ho and Lee lies not there
but in the implication that σ(t, T) = σ.
Two extra things need mentioning. First, the bond price and the cash bond price
are both lognormally distributed and thus the BlackScholes formula can still hold
(as hinted at in section 4.1, and shown in section 6.2).
And second, there is a straightforward generalization to a deterministic shortrate
volatility
dr
t
= σ
t
dW
t
+ θ
t
dt,
with a corresponding HJM formulation
d
t
f(t, T) = σ
t
dW
t
+ σ
2
t
(T −t)dt,
with the initial forward rate curve given by
f(0, T) = r
0
−
_
T
0
σ
2
s
(T −s)ds +
_
T
0
θ
s
ds.
120 CHAPTER 5. INTEREST RATES
The extra freedom here is to allow the volatility surface to depend on time, but not
on maturity. For that we require something else...
Vasicek/HullWhite
Next in the accepted hierarchy is to allow the short rate’s drift to depend on its current
value.
Vasicek model
We model the short rate (under Q) as:
dr
t
= σdW
t
+ (θ −αr
t
)dt
for some constant α, θ and σ.
The SDE is composed of a Brownian part and a restoring drift which pushes it
upwards when the process is below θ/α and downwards when it is above. The mag
nitude of the drift is also proportional to the distance away from this mean. Such a
process is called an OrnsteinUhlenbeck or OU process.
We can use It ˆ o’s formula to check that the solution to this, starting r at r
0
, is
r
t
= θ/α + e
−αt
(r
0
−θ/α) + σe
−αt
_
t
0
e
αs
dW
s
.
As it happens, r
t
can be rewritten in terms of a different QBrownian motion
¯
W as
r
t
= θ/α + e
−αt
(r
0
−θ/α) + σe
−αt
¯
W
_
e
2αt
−1
2α
_
,
so that r
t
has a normal marginal distribution with mean θ/α + exp(−αt)(r
0
− θ/α)
and variance σ
2
(1 −exp(−2αt))/2α. As t gets large, this converges to an equilibrium
normal distribution of mean θ/α and variance σ
2
/2α. This does not mean that the
process r
t
converges — it doesn’t — only that its distribution converges.
Figure 5.7: An OU process, with σ = 2θ = 2α = 1
What HJM model are we in? Again we can use It ˆ o to ﬁnd g(x, t, T), and thus
σ(t, T) and f(0, T). In this case
5.4. SHORTRATE MODELS 121
Vasicek model in HJM terms
σ(t, T) = σ exp
_
−α(T −t)
_
,
with f(0, T) = θ/α + e
−αT
(r
0
−θ/α) −
σ
2
2α
2
(1 −e
−αT
)
2
.
Now we can see an advantage over Ho and Lee — where Ho and Lee failed to
introduce a maturity dependence into the volatility surface, this model can. Thus this
model is capable of calibration to a richer set of observed volatilities. Note how the
volatility σ(t, T) is derived from both the drift and volatility of the short rate under
Q. In order to describe an HJM model, we need two degrees of freedom for the
volatility — one for time and one for maturity. The short rate description doesn’t
abandon the second degree of freedom; it encodes it in the relationship between its
volatility ρ(r
t
, t) and its drift ν(r
t
, t). The drift of r
t
under Q is a vital part of the
description.
But only under Q. The Vasicek model, unlike Ho and Lee, may be mean reverting
under Q, but both models are in fact capable of mean reversion under P. Some care
has to be taken — the introduction of the extra parameter α does give Vasicek a richer
set of allowable Pdrifts than Ho and Lee, but this richness involves maturity. Sim
ple timedependent considerations will not in general prejudice one over the other.
Because it is possible to ﬁnd a change of measure γ
t
which gives mean reverting be
havior to Ho and Lee, Vasicek is not the inevitable choice if in the real world mean
reversion is observed. In practice it will be the volatility of the entire curve, rather
than the drift of the short rate that forces one over the other.
As before there is a natural generalization to
dr
t
= σ
t
dW
t
+ (θ
t
−α
t
r
t
)dt
where σ
t
, θ
t
, and α
t
are deterministic functions of time. As r
t
is still a Gaussian
process with normal marginals, so f(t, T) is Gaussian and the bond prices have log
normal marginals. In this case, the HJM volatility and initial forward curve are
σ(t, T) = σ
t
β(t, T), where β(t, T) = exp
_
−
_
T
t
α
s
ds
_
, and
f(0, T) = r
0
β(0, T) +
_
T
0
θ
s
β(s, T)ds −
_
T
0
σ
2
s
β(s, T)
_
_
T
s
β(s, u)du
_
ds.
The normality of the forward rates f(t, T) is both good news and bad news. In
its favor, it means that the bond prices P(t, T) are lognormally distributed, so that
the lognormal option pricing results of section 6.2 all hold. On the other hand,
both the instantaneous rate and the forward rates can go negative from time to time.
Depending on the parameters, this can happen more or less rarely — the next model
rectiﬁes this defect.
122 CHAPTER 5. INTEREST RATES
CoxIngersollRoss
The model is a meanreverting process, which pushes away from zero to keep it pos
itive (see box).
CoxIngersollRoss model
The instantaneous rate’s SDE, under Q, is
dr
t
= σ
t
√
r
t
dW
t
+ (θ
t
−α
t
r
t
)dt,
where σ
t
, θ
t
, and α
t
are deterministic functions of time.
The drift term is a restoring force which always points towards the current mean
value of θ
t
/α
t
. The volatility term is set up to get smaller as r
t
approaches zero, so
allowing the drift θ
t
to dominate and to stop r
t
from going below zero.
As long as θ satisﬁes θ
t
≥
1
2
σ
2
t
, then the process actually stays strictly positive.
This process is called autoregressive.
Figure 5.8: Autoregressive σ = 1, θ = 2, α = 2
It is difﬁcult to ﬁnd an explicit pathwise solution for r
t
, but we can solve a useful
partial differential equation (PDE) . Firstly deﬁne B(t, T) to be the solution of the
Riccati differential equation
∂B
∂t
=
1
2
σ
2
t
B
2
(t, T) + α
t
B(t, T) −1, B(T, T) = 0.
(In general, this equation has no analytic solution, but it has been well studied nu
merically.) Then the function g(x, t, T), which is (−log P(t, T)r
t
= x) can be written
in terms of this solution B(t, T) as
g(x, t, T) = xB(t, T) +
_
T
t
θ
s
B(s, T)ds.
Letting D(t, T) be
∂B
∂T
(t, T), then the volatility structure can be expressed as
5.5. MULTIFACTOR HJM 123
CoxIngersollRoss in HJM terms
σ(t, T) = σ
t
√
r
t
D(t, T),
and Σ(t, T) = −σ
t
√
r
t
B(t, T),
with f(0, T) = r
0
D(0, T) +
_
T
0
θ
s
D(s, T)ds.
As usual, the bond price P(t, T) has the form P(t, T) = exp −g(r
t
, t, T).
BlackKarasinski
Another way round the problem of keeping the short rate positive is to take expo
nentials. This model is an extension of the BlackDermanToy model and starts by
taking X
t
to be the general OU process of the Vasicek model. Explicitly:
BlackKarasinski model
The process X
t
is
dX
t
= σ
t
dW
t
+ (θ
t
−α
t
X
t
)dt,
where σ
t
, θ
t
, and α
t
are deterministic functions of time. The instantaneous rate r
t
is then assumed to be
r
t
= exp(X
t
).
So the logarithm of the rate drifts towards the current mean of θ
t
/α
t
. The rate itself
also drifts towards a mean, and additionally is always positive. We also know that
X
t
is normally distributed, so that r
t
is lognormal. However,
_
T
t
r
s
ds is awkward
to examine analytically. This model is still HJM consistent; that is, there is some
volatility surface σ(t, T) which generates a singlefactor HJM model which has the
same instantaneous rate as given above.
5.5 Multifactor HJM
The drawback of the singlefactor model is that all the increments in the bond prices
are perfectly correlated. For many applications, that assumption is too coarse, es
pecially if we are trying to price something which depends on the difference of two
points on the yield curve.
A multifactor model involves driving the various processes by a collection of in
dependent Brownian motions. More details of such models are in section 6.3. In an
nfactor model, we will have n Brownian motions to work with: W
1
(t), . . . , W
n
(t).
Correspondingly each Tbond forward rate process has a volatility σ
i
(t, T) for each
Brownian factor W
i
(t). This allows different bonds to depend on external ‘shocks’ in
different ways, and to have strong correlations with some bonds and weaker correla
124 CHAPTER 5. INTEREST RATES
tions with others. The general form of the multifactor HJM model is
f(t, T) = f(0, T) +
n
i=1
_
t
0
σ
i
(s, T)dW
i
(s) +
_
t
0
α(s, T)ds, 0 ≤ t ≤ T,
which is to say that the forward process starts with initial value f(0, T) and is driven
by various Brownian terms and a drift. From this, the total instantaneous square
volatility of f(t, T), and and the covariance of the increments of the two forward
rates f(t, T) and f(t, S) are respectively
n
i=1
σ
2
i
(t, T), and
n
i=1
σ
i
(t, T)σ
i
(t, S).
In the singlefactor model, n is 1, and the correlation of the changes in the forward
rates of the Tbond and Sbond is exactly one.
The instantaneous rate r
t
= f(t, t) can be written, similarly to before as
r
t
= f(0, t) +
n
i=1
_
t
0
σ
i
(s, t)dW
i
(s) +
_
t
0
α(s, t)ds.
The volatility and drift conditions are generalized to:
Multifactor HJM: conditions on the volatilities and drift
We assume that
• for each T, the processes σ
i
(t, T) and α(t, T) Fprevisible and their integrals
_
T
0
σ
2
i
(t, T)dt and
_
T
0
α(t, T)dt are ﬁnite;
• the initial forward curve, f(0, T), is deterministic and satisﬁes the condition
that
_
T
0
f(0, u)du < ∞;
• the drift α has ﬁnite integral
_
T
0
_
u
0
α(t, u)dtdu;
• each volatility σ
i
has ﬁnite expectation E
_
T
0
¸
¸
_
u
0
σ
i
(t, u)dW
i
(t)
¸
¸
du.
To make the discounted bond prices into martingales, we need a version of the
CameronMartinGirsanov theorem for higher dimensions (section 6.3). The con
ditions we need for this to work are shown in the two boxes, where Σ
i
(t, T) is the
integral −
_
T
t
σ
i
(t, u)du.
Multifactor HJM: market completeness conditions (1)
It is required that
• there exist previsible processes γ
i
(t), for 1 ≤ i ≤ n, such that
• the expectation Eexp
1
2
_
T
0
γ
2
i
(t)dt is ﬁnite;
This is just as in the singlefactor case, but with one difference. The drift is now
allowed n ‘dimensions of freedom’ away from its riskneutral value. That is, as a
5.5. MULTIFACTOR HJM 125
function of T, α(t, ·) is allowed to deviate by any linear combination of the functions
α
i
(t, ·). This is still much less than the set of all possible functions, but it is larger
than in the singlefactor case. The second condition is the technical requirement of
the CMG theorem for γ
i
(t) to be a drift under an equivalent change of measure.
Multifactor HJM: market completeness conditions (2)
We also need that
• the matrix A
t
= (Σ
i
(t, T
j
))
n
i,j=1
is nonsingular for almost all (t, ω), t < T
1
, for
every set of maturities T
1
< T
2
< · · · < T
n
;
• and the expectation Eexp
1
2
n
i
_
T
0
_
γ
i
(t) −Σ
i
(t, T)
_
2
dt is ﬁnite.
The modiﬁcation from the singlefactor case here is that the volatility process
A
t
which used to be required to be nonzero has been replaced by a volatility matrix
process which has to be nonsingular. The second condition ensures that the resulting
driftless discounted bond price is in fact a martingale (a multidimensional equivalent
of the collector’s guide to exponential martingales).
As before we ﬁnd that the bond prices themselves have stochastic increments
d
t
P (t, T) = P (t, T)
_
n
i=1
Σ
i
(t, T) dW
i
(t)
+
_
r
t
−
_
T
t
_
α(t, u) +
n
i=1
σ
i
(t, u) Σ
i
(t, u)
_
du
_
dt
_
,
where Σ
i
(t, T) is the integral −
_
T
t
σ
i
(t, u)du. The discounted bond prices Z(t, T) =
B
−1
t
P(t, T) satisfy
d
t
Z (t, T) = Z (t, T)
_
n
i=1
Σ
i
(t, T) dW
i
(t)
−
_
_
T
t
_
α(t, u) +
i
σ
i
(t, u) Σ
i
(t, u)
_
du
_
dt
_
.
The SDE for Z now becomes
d
t
Z(t, T) = Z(t, T)
n
i=1
Σ
i
(t, T)
_
dW
i
(t) + γ
i
(t)dt
_
.
Using the multidimensional CMG, we can ﬁnd a measure Q equivalent to P, under
which
˜
W
1
, . . . ,
˜
W
n
are independent QBrownian motions, where
˜
W
i
(t) = W
i
(t) +
_
t
0
γ
i
(s)ds. So Z’s SDE is (in Qterms)
d
t
Z(t, T) = Z(t, T)
n
i=1
Σ
i
(t, T)d
˜
W
i
(t),
and every Z(t, T) is a Qmartingale in t.
126 CHAPTER 5. INTEREST RATES
Under this martingale measure, the bond price P and the forward rate f have the
stochastic differentials
Bond prices and forward rates under Q
d
t
P(t, T) = P(t, T)
_
n
i=1
Σ
i
(t, T)d
˜
W
i
(t) + r
t
dt
_
,
d
t
f(t, T) =
n
i=1
σ
i
(t, T)d
˜
W
i
(t) −
n
i=1
σ
i
(t, T)Σ
i
(t, T)dt.
Derivative pricing and hedging
The actual price of a derivative still has a familiar form:
Option price formula (HJM)
If X is the payoff of a derivative at time T, then its value at time t is
V
t
= B
t
E
Q
_
B
−1
T
XF
t
_
= E
Q
_
exp
_
−
_
T
t
r
s
ds
_
X
¸
¸
¸
¸
¸
F
t
_
.
We also need a multidimensional martingale representation theorem. Formally
Martingale representation theorem (nfactor)
Let
˜
W be ndimensional QBrownian motion, and suppose that M
t
is an n
dimensional Qmartingale process, M
t
=
_
M
1
(t), . . . , M
n
(t)
_
, which has volatility
matrix
_
σ
ij
(t)
_
, in that dM
j
(t) =
i
σ
ij
(t)d
˜
W
i
(t), and the matrix satisﬁes the ad
ditional condition that (with probability one) it is always nonsingular. Then if N
t
is any onedimensional Qmartingale, there exists an ndimensional Fprevisible
process φ
t
=
_
φ
1
(t), . . . , φ
n
(t)
_
such that
_
T
0
(
i
σ
ij
(t)φ
j
(t))
2
dt < ∞, and the mar
tingale N can be written as
N
t
= N
0
+
n
j=1
_
t
0
φ
j
(s)dM
j
(s).
Further φ is (essentially) unique.
As a general rule, if we have an nfactor model, we need a trading portfolio of n
separate instruments, as well as the cash bond, in order to hedge claims. An advan
tage of the HJM framework is that we are free to choose whichever n instruments we
like, and the answer will always be the same.
If we are going to hedge the claim X with discount bonds, we must still make
sure that all their maturities are later than T. Suppose we choose to use bonds with
maturities T
1
, T
2
, . . . , T
n
all larger than T.
5.6. INTEREST RATE PRODUCTS 127
A selfﬁnancing strategy
_
φ
1
(t), . . . , φ
n
(t), ψ
t
_
will be the combination of both an
nvector of holdings in the bonds with maturities T
1
, . . . , T
n
respectively, and a hold
ing ψ
t
in the cash bond B
t
. The value of the portfolio at time t is
V
t
=
n
j=1
φ
j
(t)P(t, T
j
) + ψ
t
B
t
,
and its discounted value E
t
= B
−1
t
V
t
is
E
t
=
n
j=1
φ
j
(t)Z(t, T
j
) + ψ
t
.
The selfﬁnancing equality for the strategy (as in section 6.4) is that
dE
t
=
n
j=1
φ
j
(t)d
t
Z(t, T
j
).
We can now apply the representation theorem in the usual way to the martingale
produced from the discounted claim, that is E
t
= E
Q
(B
−1
T
XF
t
). The part of the
martingales M
j
(t) will be taken by the discounted bonds Z(t, T
j
). Their volatility
matrix is given by A
t
=
_
Σ
i
(t, T
j
)
_
i,j
, which is nonsingular by the completeness
conditions box. If we set E
t
= E
Q
(B
−1
T
XF
t
), then by the representation theorem,
there is an nvector of previsible processes φ
t
such that
E
t
= E
Q
(B
−1
T
X) +
n
j=1
_
t
0
φ
j
(s)dZ(s, T
j
).
This immediately gives a selfﬁnancing strategy φ. We hold φ
j
(t) units of the T
j
bond
at time t, and ψ
t
= E
t
−
j
φ
j
(t)Z(t, T
j
) units of the cash bond.
In the usual way, the portfolio costs an initial E
Q
(B
−1
T
X) and evolves to be worth
exactly X by time T.
5.6 Interest rate products
In recent years, there has been a great increase in the number of interest rate products
available. Especially in the overthecounter markets, contracts which not long ago
would have been considered as exotics are now commonplace. We cannot hope to
describe the hundreds and possibly thousands of traded claims, but we can sketch out
the basic types within each area.
Forward contract
This is about the simplest product possible. We agree, at the current time t, to make
a payment of an amount k at a future time T
1
, and in return to receive a dollar at the
later time T
2
. What should the amount k be?
128 CHAPTER 5. INTEREST RATES
According to the pricing formula (under whatever model we are in), the value of
the claim now is
V
t
= B
t
E
Q
_
B
−1
T
2
F
t
_
−B
t
E
Q
_
kB
−1
T
1
F
t
_
,
under the martingale measure Q, where B
t
is the cash bond
B
t
= exp
_
t
0
r
s
ds.
Recalling that B
t
E
Q
_
B
−1
T
F
t
_
is just P(t, T), we see that
V
t
= P(t, T
2
) −kP(t, T
1
).
For this contract to have null current net worth, we merely choose k at time t to be
k =
P(t, T
2
)
P(t, T
1
)
.
This price makes sense, as saying that the forward yield from T
1
to T
2
is
−
log P(t, T
2
) −log P(t, T
1
)
T
2
−T
1
.
For T
1
and T
2
very close together, this approximates to the instantaneous forward rate
of borrowing
−
∂
∂T
log P(t, T) = f(t, T).
The price also gives us a clue to the hedging strategy. Suppose we were, at time t, to
buy k units of the T
1
bond and sell one unit of the T
2
bond. The initial cost of that
deal is zero, and the portfolio pays us k at time T
1
(matching the payment we have to
make at that time) and exactly absorbs the dollar we receive at time T
2
.
In this particular example, the answer is independent of our particular term struc
ture model, as the hedging strategy is static. There are other important cases where
this also happens.
Multiple payment contracts
Most interest rate products don’t just make a single payment X at time T. Instead
the contract speciﬁes a sequence of payments X
i
made at a sequence of times T
i
(i = 1, . . . , n). Each payment X
i
may depend on price movements up to its payment
time T
i
, and even on any previous payment. As long as we bear that in mind, this
causes no serious problem, and indeed there are two different ways to keep things
clear.
• Divide and rule. We can treat each payment X
i
separately. On its own, it is just
a claim at time T
i
, so its worth at time t is
V
i
(t) = B
t
E
Q
(B
−1
T
i
X
i
F
t
) = P(t, T
i
)E
P
T
i
(XF
t
),
where P
T
i
, is the T
i
forward measure (see section 6.4). This approach will al
ways work, but the forward measure, if used, will have to be changed for each
i.
5.6. INTEREST RATE PRODUCTS 129
• Savings account. We could instead roll up the payments into savings as we get
them, and keep them till the last payment date T. That is, as each payment is
made, we use it to buy a Tbond (or invest it in the bank account process B
t
till
time T). Then the payoff is a single payment at time T of
X =
n
i=1
X
i
P(T
i
, T)
with worth at time t
V
t
= B
t
E
Q
(B
−1
T
XF
t
) = P(t, T)E
P
T
(XF
t
).
Bonds with coupons
In practice, pure discount bonds with no coupon are not popular products. Especially
at the long end. Instead, a bond may not only pay its principal back at maturity, but
also make smaller regular coupon payments of a ﬁxed amount c up until then.
Suppose a bond makes n regular payments at (uncompounded) rate k at times
T
i
= T
0
+ iδ (i = 1, . . . , n) and also pays off a dollar at time T
n
. The amount of the
actual coupon payment is kδ, where δ is the payment period. This income stream
is equivalent to owning one T
n
bond and kδ units of each T
i
bond. The price of the
coupon bond at time T
0
is
P(T
0
, T
n
) + kδ
n
i=1
P(T
0
, T
i
).
If we desire the bond to start with its face value, then the coupon rate should be
k =
1 −P(T
0
, T
n
)
δ
n
i=1
P(T
0
, T
i
)
.
Floating rate bonds
A bond might also pay off a coupon which was not ﬁxed, but depended on current
interest rates. One interesting case is where the interest paid over an interval from
time S to time T is the same as the yield of the Tbond bought at time S.
Suppose a bond pays its dollar principal at time T
n
, and also payments at times
T
i
= T
0
+iδ (i = 1, . . . , n) of varying amounts. The amount of payment made at time
T
i
is determined by the LIBOR rate set at time T
i−1
L(T
i−1
) =
1
δ
_
1
P(T
i−1
, T
i
)
−1
_
.
The actual payment made at time T
i
is δL(T
i−1
) = P(T
i−1
, T
i
)
−1
− 1, which is the
amount of interest we would receive by buying a dollar’s worth of the T
i
bond at time
T
i−1
.
The value to us now, at time T
0
, of the T
i
payment is
B
T
0
E
Q
_
B
−1
T
i
(P
−1
(T
i−1
, T
i
) −1)F
T
0
_
.
130 CHAPTER 5. INTEREST RATES
Because the conditional expectation E
Q
(B
−1
T
i
F
T
i−1
) is B
−1
T
i−1
P(T
i−1
, T
i
) and the bond
price P(T
i−1
, T
i
) is known with respect to the F
T
i−1
information, we can divide it
through both sides to get
E
Q
_
B
−1
T
i
P
−1
(T
i−1
, T
i
)
¸
¸
F
T
i−1
_
= B
−1
T
i−1
.
Using the tower law, we can rewrite the value of the T
i
payment as
B
T
0
E
Q
_
B
−1
T
i−1
−B
−1
T
i
¸
¸
¸ F
T
0
_
,
which is just P(T
0
, T
i−1
) − P(T
0
, T
i
). This price also suggests the hedge of selling a
T
i
bond and buying a T
i−1
bond. When the T
i−1
bond matures, we buy P
−1
(T
i−1
, T
i
)
units of the T
i
bond, and we are left with exactly the right payoff at time T
i
.
The total value of the variable coupon bond is the sum of its components. That is,
V
0
= P(T
0
, T
n
) +
n
i=1
_
P(T
0
, T
i−1
) −P(T
0
, T
i
)
_
= 1.
Surprisingly, the bond has a ﬁxed price equal to the face value of its principal. Why
this is so, is because the bond is equivalent to this simple sequence of trades:
• take a dollar and buy T
1
bonds with it
• take the interest from the bonds at T
1
as a coupon, and buy some T
2
bonds with
the dollar principal
• repeat until we are left with the dollar at time T
n
.
This has exactly the same cash ﬂows as the variable coupon bond, so the initial
prices must match.
Swaps
This very popular contract simply exchanges a stream of varying payments for a
stream of ﬁxed amount payments (or vice versa). That is, we swap a ﬂoating interest
rate for a ﬁxed one.
Typically, we might offer a contract where we receive a regular sequence of ﬁxed
amounts and at each payment date we pay an amount depending on prevailing interest
rates. In practice, only the net difference is exchanged, as shown in ﬁgure 5.9:
A standard deﬁnition of the variable payment is that of the interest paid by a bond
over the previous time period. If the payment dates are T
i
= T
0
+ iδ (i = 1, . . . , n),
then the ith payment will be determined by the δperiod LIBOR rate set at time T
i−1
.
The payment made is
δL(T
i−1
) =
1
P(T
i−1
−T
i
)
−1.
Suppose the swap pays at a ﬁxed rate k at each time period. Then the swap looks like
a portfolio which is long a ﬁxed coupon bond and short a variable coupon bond. We
5.6. INTEREST RATE PRODUCTS 131
(a) Gross payments received and given (b) Net receipts
Figure 5.9:
know that the former is worth
P(T
0
, T
n
) + kδ
n
i=1
P(T
0
, T
i
),
and the latter costs a dollar. The ﬁxed rate needed to give the swap initial null value
is
k =
1 −P(T
0
, T
n
)
δ
n
i=1
P(T
0
, T
i
)
.
Forward swaps
In a forward swap agreement, we have chosen to receive ﬁxed payments at rate k,
starting at time T
0
with payments at times T
i
= T
0
+ iδ (i = 1, . . . , n). The value of
this swap at time T
0
will be
X = P(T
0
, T
n
) + kδ
n
i=1
P(T
0
, T
i
) −1.
The present value of X at time t before T
0
is given by the formula:
V
t
= B
t
E
Q
(B
−1
T
0
XF
t
) = P(t, T
n
) + kδ
n
i=1
P(t, T
i
) −P(t, T
0
).
The ﬁxed rate needed to give the forward swap initial null value at time t is
k =
P(t, T
0
) −P(t, T
n
)
δ
n
i=1
P(t, T
i
)
.
This rate k is the forward swap rate. An alternative formulation of this expression is
k =
1 −F
t
(T
0
, T
n
)
δ
n
i=1
F
t
(T
0
, T
i
)
,
where F
t
(T
0
, T
i
) is the forward price at time t for purchasing a T
i
bond at time T
0
That is F
t
(T
0
, T
i
) = P(t, T
i
)/P(t, T
0
). In this form the expression resembles the in
stantaneous swap rate.
132 CHAPTER 5. INTEREST RATES
Bond options
Like a stock option, a bond option gives the right to buy a bond at a future date for
a given price. An option on a Tbond, struck at k with exercise time t, has current
worth
E
Q
_
B
−1
t
_
P(t, T) −k
_
+
_
,
where Q is the martingale measure.
Under the Ho and Lee model, where the forward rates evolve as
d
t
f(t, T) = σdW
t
+ σ
2
(T −t)dt,
the forward rates and the instantaneous short rate are normally distributed. This
makes the Tbond and the discount bond lognormally distributed, so that we can
price the option with the lognormal results of section 6.2. The option price is
V
0
= P (0, t)
_
FΦ
_
log
F
k
+
1
2
¯ σ
2
t
¯ σ
√
t
_
−kΦ
_
log
F
k
−
1
2
¯ σ
2
t
¯ σ
√
t
__
,
where F is the current forward price for P(t, T), that is F = P(0, T)/P(0, t), and
the term volatility ¯ σ is σ(T − t) (that is, ¯ σ
2
t is the logvariance of P(t, T)). Under
the Vasicek model, which is the most general singlefactor model with lognormal
bond prices, this formula also holds with the same forward price, but a different ¯ σ
depending on the deterministic processes σ
t
and φ
t
in the model.
Compare this with the price of an option on a stock S, with volatility σ, struck at
price k with exercise time t. It is worth
V
0
= e
−rt
_
FΦ
_
log
F
k
+
1
2
σ
2
t
σ
√
t
_
−kΦ
_
log
F
k
−
1
2
σ
2
t
σ
√
t
__
,
where r is the constant interest rate and F is the current forward price of the stock,
that is F = e
rt
S
0
, and σ is the (term) volatility of S
t
.
We see that the bond option price formula merely changes the discount factor rep
resenting the value now of a dollar at time t. Under constant interest rates this was
e
rt
, and under variable interest rates it is just the price of a tbond P(0, t). Other
wise, as long as the other variables are expressed in terms of forward prices and term
volatilities, the formula is the same.
Options on coupon bonds
Imagine a bond which pays coupons at rate k at the times T
i
= T
0
+ iδ (i = 1, . . . , n)
before redeeming a dollar at time T
n
. We can buy or sell the bond before time T
n
,
transferring the ownership of future (but not past) coupons along with it. As we’ve
seen before the value of this bond at time t is
C
t
= P(t, T
n
) + kδ
n
i=I(t)
P(t, T
i
),
5.6. INTEREST RATE PRODUCTS 133
where I(t) = min{i : t < T
i
} is the sequence number of the next coupon payment
after time t.
Suppose we have an option to buy the bond at time t for price K. In general it
is not easy to value this option analytically. However, in the special case where we
have a singlefactor model with a Markovian short rate, we can price the option more
easily using a trick of Jamshidian.
Each bond price P(t, T) can be seen as a deterministic function P(t, T; r
t
) of time,
maturity and the instantaneous rate. Additionally, this function will be decreasing in
r
t
— as rates rise, prices fall. A portfolio which is long a number of bonds will have
the same behavior. So C
t
itself will be a function C(t; r
t
) which is decreasing in r
t
.
Thus there is some critical value r
∗
of r such that C(t; r
∗
) is exactly K. Setting
K
i
to be P(t, T
i
; r
∗
), then r
∗
is also critical for an option on the T
i
bond struck at K
i
.
This means that C
t
is larger than K if and only if any (and every) P(t, T
i
) is larger
than K
i
. And so
(C
t
−K)
+
=
_
P(t, T
n
) −K
n
_
+
+ kδ
n
i=I(t)
_
P(t, T
i
) −K
i
_
+
.
In other words, an option on this portfolio is a portfolio of options, and we can price
each one using the zerocoupon bond option formula.
Caps and ﬂoors
Suppose we are borrowing at a ﬂoating rate and want to insure against interest pay
ments going too high. If we make payments at times T
i
= T
0
+ iδ (i = 1, . . . , n), then
we pay at time T
i
the δperiod LIBOR rate set at time T
i−1
L(T
i−1
) =
1
δ
_
1
P(T
i−1
−T
i
)
−1
_
.
How much would it cost to ensure that this rate is never greater than some ﬁxed rate
k? The cap contract pays us the difference between the LIBOR and the cap rate
δ
_
L(T
i−1
) −k
_
+
at each time T
i
. An individual payment at a particular time T
i
is called a caplet, and
if we can price caplets, we can price the cap.
Now we can rewrite the caplet claim as
X = (1 + kδ)P
−1
i
(K −P
i
)
+
,
where P
i
is P(T
i−1
, T
i
) and K is (1 + kδ)
−1
. The value of the caplet at time t is
B
t
E
Q
(B
−1
T
i
X mcF
t
), which equals
(1 +kδ)B
t
E
Q
_
B
−1
T
i−1
(K −P
i
)
+
¸
¸
¸ F
t
_
.
This is just equal to the value of (1 + kδ) put options on the T
i
bond, struck at K,
exercised at T
i−1
. The option price formula (and putcall parity) will then price the
caplet.
134 CHAPTER 5. INTEREST RATES
A ﬂoor works similarly, but inversely, in that we receive a premium for agreeing
to never pay less than rate k at each time T
i
. That is, we pay an extra amount
δ
_
k −L(T
i−1
)
_
+
at time T
i
. There is a ﬂoorcap parity which says that the worth of a ‘ﬂoorlet’ less the
cost of a caplet equals (1 + kδ)P(t, T
i
) −P(t, T
i−1
). Buying a ﬂoor and selling a cap
at the same strike k is exactly equivalent to receiving ﬁxed at rate k on a swap.
Swaptions
A swaption is an option to enter into a swap on a future date at a given rate. Suppose
we have an option to receive ﬁxed on a swap starting at date T
0
. The swap payment
dates are T
i
= T
0
+ iδ (i = 1, . . . , n), and the ﬁxed swap rate is k. Then the worth of
the option at time T
0
is
_
P (T
0
, T
n
) + kδ
n
i=1
P (T
0
, T
i
) −1
_
+
.
This is exactly the same as a call option, struck at 1, on a T
n
bond which pays a
coupon at rate k at each time T
i
. That is not entirely a coincidence as a swap is just a
coupon bond less a ﬂoating bond (which always has par value). If you receive ﬁxed
on a swap, you have a long position in the bond market; a swap option looks like a
bond option.
5.7 Multifactor models
If we want to price a product depending on a range of bonds, it makes more sense to
use a multifactor model. A simple case is given in HeathJarrowMorton’s original
paper. It is an extension of Ho and Lee’s model to two factors.
A twofactor model
Suppose the forward rates evolve as
d
t
f(t, T) = σ
1
dW
1
(t) + σ
2
e
−λ(T−t)
dW
2
(t) + α(t, T)dt,
where σ
1
, σ
2
and λ are constants, and α is a deterministic function of t and T. Here
the W
1
Brownian motion provides ‘shocks’ which are felt equally by points of all
maturities on the yield curve, whereas W
2
gives shortterm shocks which have little
effect on the longterm end of the curve. This model is HJM consistent, so we can
read off information about it from that structure. The HJM completeness conditions
reduce, in this case, to there being two Fprevisible processes γ
1
(t) and γ
2
(t) such
that the drift α is
α(t, T) = σ
1
γ
1
(t) + σ
2
e
−λ(T−t)
γ
2
(t) + σ
2
1
(T −t) +
σ
2
2
λ
_
1 −e
−λ(T−t)
_
e
−λ(T−t)
.
5.7. MULTIFACTOR MODELS 135
So the range of available drifts has two degrees of functional freedom away from the
martingale measure drift. Under the martingale measure (that is γ
1
= γ
2
= 0), the
forward rate is
f (t, T) = σ
1
W
1
(t) + σ
2
e
−λT
_
t
0
e
λs
dW
2
(s) + f (0, T) +
_
t
0
α(s, T) ds.
Like Ho and Lee, this model has normally distributed forward rates — which does
allow them to go negative. Nevertheless the model does have the advantages of tech
nical tractability and an explicit option formula. We can deduce from the forward
rate formula that −log P (t, T) =
_
T
t
f (t, u) du is
σ
1
(T −t) W
1
(t)+
σ
2
λ
_
e
−λt
−e
−λT
_
_
t
0
e
λs
dW
2
(s) +
_
T
t
f (0, u) du +
_
t
0
_
T
t
α(s, u) duds,
This means that the instantaneous rate is made up of a Brownian motion and an in
dependent meanreverting (OrnsteinUhlenbeck) process plus drift. However in a
multifactor setting, the short rate loses its dominant role as the carrier of all infor
mation about the bond prices.
Setting ¯ σ
2
(t, T) to be the variance (term variance) of log P(t, T), we have
¯ σ
2
(t, T) = σ
2
1
(T −t)
2
t +
_
σ
2
λ
_
1 −e
−λ(T−t)
__
2
1
2λ
_
1 −e
−2λt
_
.
The discounted bond, B
t
= exp
_
_
t
0
r
s
ds
_
, is also lognormally distributed, because
we can deduce that the integral
_
t
0
r
s
ds is normal from the expression for r
t
above.
We can use the results of section 6.2, given the joint lognormality of the asset and
discount bond prices. The value of an option on the Tbond, struck at k, exercised at
time t is
V
0
= P (0, t)
_
FΦ
_
log
F
k
+
1
2
¯ σ
2
(t, T)
¯ σ (t, T)
_
−kΦ
_
log
F
k
−
1
2
¯ σ
2
(t, T)
¯ σ (t, T)
__
,
where F is P(0, T)/P(0, t), the forward price of the Tbond. This BlackScholes type
of formula allows us to price caps and ﬂoors as well as options on the discount T
bonds. However, in the multifactor setting, the trick we used before to price options
on couponbearing bonds does not work, making it more involved to price them and
the associated swaptions.
The general multifactor normal model
We can actually generalize the twofactor model above to a general multifactor one
which also has normal forward rates and an explicit BlackScholes type option pric
ing formula.
We take the instance of the completely general nfactor model, where each volatil
ity surface σ
i
(t, T) can be written as a product
σ
i
(t, T) = x
i
(t)y
i
(T),
136 CHAPTER 5. INTEREST RATES
where x
i
and y
i
are deterministic functions. The forward rates are then driven by
d
t
f(t, T) =
n
i=1
y
i
(T)x
i
(t)dW
i
(t) + α(t, T)dt.
Here the function x
i
determines the size at time t of ‘type i shocks’, and the function
y
i
controls how the shock is felt at different maturities. In the singlefactor case when
n = 1, this framework incorporates both the Ho and Lee model (x(t) = σ, y(T) = 1)
and the Vasicek model (x (t) = σ
t
exp
_
_
t
0
α
s
ds
_
, y (T) = exp
_
−
_
T
0
α
s
ds
_
).
For the market to be complete, we need two conditions on the functions α and y
i
to hold. Firstly, there should be n Fprevisible processes γ
1
, . . . , γ
n
, such that
α(t, T) =
n
i=1
x
i
(t) y
i
(t)
_
γ
i
(t) + x
i
(t) Y
i
(t, T)
_
,
where Y
i
(t, T) =
_
T
t
y
i
(u)du. In other words, the drifts consistent with hedging span
an ndimensional function space around the martingale drift. Secondly the matrix
A
t
(a
ij
(t)), where a
ij
(t) = Y
j
(t, T
i
) should be nonsingular for all t < T
1
, for every
set of n maturities T
1
< · · · < T
n
. This condition is really just asserting that all the
functions y
i
are different. It is satisﬁed, for instance, if each volatility σ
i
has the form
σ
i
(t, T) = σ
i
(t) exp
_
−λ
i
(T −t)
_
,
where the σ
i
(t) are deterministic functions of time and the λ
i
are distinct constants.
For the general volatility surface σ
i
(t, T) = x
i
(t)y
i
(T), the short rate and the for
ward rates are normally distributed. Consequently the bond prices are lognormally
distributed and a BlackScholes type formula holds (see section 6.2). Let F be the
forward price of the Tbond at time t, F = P(0, T)/P(0, t), and let σ be the term
volatility of the Tbond up to time t, that is σ
2
t is the variance of log P(t, T), or
σ
2
=
1
t
n
i=1
Y
2
i
(t, T)
_
t
0
x
2
i
(s)ds.
Then the value at time zero of a call on the Tbond, struck at k, exercisable at time t
is
V
0
= P (0, t)
_
FΦ
_
log
F
k
+
1
2
σ
2
t
σ
√
t
_
−kΦ
_
log
F
k
−
1
2
σ
2
t
σ
√
t
__
,
BraceGatarekMusiela
The BraceGatarekMusiela (BGM) model is a particular case of HJM which focuses
on the δperiod LIBOR rates. We shall simplify their notation slightly and write
L(t, T) =
1
δ
_
P (t, T)
P (t, T + δ)
−1
_
.
So L(t, T) is the δperiod (forward) LIBOR rate for borrowing at a time T.
5.7. MULTIFACTOR MODELS 137
The general HJM model (of n factors) deﬁned by the forward volatilities σ
i
(t, T)
is restricted in the BGM setup to those σ such that
_
T+δ
T
σ
i
(t, u) du =
δL(t, T)
1 + δL(t, T)
γ
i
(t, T)
holds for all t less than T. Here, γ is some deterministic R
n
valued function which is
absolutely continuous with respect to T.
Then it follows that, under the martingale measure Q, L obeys the SDE
d
t
L(t, T) = L(t, T)
n
i=1
γ
i
(t, T)
_
dW
i
(t) +
_
_
T+δ
t
σ
i
(t, u) du
_
dt
_
.
More interestingly, under the forward measure P
T+δ
(see section 6.4), L obeys
d
t
L(t, T) = L(t, T)
n
i=1
γ
i
(t, T) d
˜
W
i
(t),
where
˜
W
i
are P
T+δ
Brownian motions. Thus L(t, T), as a tprocess, is not only a
P
T+δ
martingale, but is also lognormal. We shall see later that this enables us to
price caps and swaptions easily.
To price, we only need to know the function γ, rather than the whole volatility
structure. While the γ function represents the correlation at time t between changes in
the LIBOR rates at different forward dates T, in practice γ is calibrated by comparing
the model’s prices with the market. For instance, in their paper, Brace, Gatarek and
Musiela ﬁt a γ function of the form
γ
i
(t, T) = f(t)γ
i
(T −t)
by calibrating against known prices of caps and swaptions.
Writing L(T) for L(T, T), the instantaneous LIBOR rate, suppose we have a con
tract which pays off at a sequence of times T
i
= T
0
+iδ (i = 1, . . . , n). If the payment
at time T
i+1
depends on the LIBOR rate set at time T
i
, for example if X = f(L(T
i
)),
then the value of that payment at time t is
V
t
= P(t, T
i+1
)E
P
T
i+1
_
f(L(T
i
))F
t
_
.
The fact that L(T
i
) is lognormally distributed under P
T
i+1
, allows us to evaluate this
expression for simple f.
One such simple f is the caplet payoff δ(L(T
i−1
) −k)
+
at time T
i
. In this case, the
worth of the caplet at time t is V
t
, equal to
δP (t, T
i
)
_
FΦ
_
log
F
k
+
1
2
ζ
2
(t, T
i−1
)
ζ (t, T
i−1
)
_
−kΦ
_
log
F
k
−
1
2
ζ
2
(t, T
i−1
)
ζ (t, T
i−1
)
__
,
where F is the forward LIBOR rate L(t, T
i−1
) and ζ
2
(t, T) is
_
T
t
γ(s, T)
2
ds, the vari
ance of log L(T) given F
t
. This valuation has the familiar BlackScholes form be
cause under the forward measure P
T
i
, L(T
i−1
) is lognormal and the calculation pro
ceeds as usual.
138 CHAPTER 5. INTEREST RATES
We can even (approximately) price swaptions. Consider the option to pay ﬁxed at
rate k and receive ﬂoating and at times T
i
= T
0
+ iδ (i = 1, . . . , n). Let us set
Γ
2
i
=
_
T
0
t
γ (s, T
i−1
)
2
ds,
which is the variance of log L(T
0
, T
i−1
) given F
t
under the forward measure P
T
i
. We
also deﬁne
d
i
=
i
j=1
δL(t, T
j−1
)
1 + δL(t, T
j−1
)
Γ
j
−
1
2
Γ
i
,
and s
0
to be the unique root of the equation
s :
n
i=1
(kδ + I (i = n))
_
_
_
i
j=1
(1 +δL(t, T
j−1
) exp (Γ
j
(s + d
j
)))
_
_
_
−1
= 1.
Then an approximation to the value at time t of the above swaption is
V
t
= δ
n
i=1
P (t, T
i
)
_
L(t, T
i−1
) Φ
_
F
i
+
1
2
Γ
2
i
Γ
i
_
−kΦ
_
F
i
−
1
2
Γ
2
i
Γ
i
__
,
where F
i
= −Γ
i
(s
0
+ d
i
).
Chapter 6
Bigger models
T
he BlackScholes stock model assumes that the stock drift and stock volatility
are constant. It assumes that there is only a single stock in the market. And
it assumes that the cash bond is deterministic with zero volatility. None of
these assumptions is necessary. The subsequent sections tackle these restrictions one
by one and show how a more general model can still price and hedge derivatives.
Also we will reveal the underlying framework which governs all these models from
behind the scenes.
This is not to say that all models, no matter how complex or bizarre, will always
give good prices. But if a model is driven by Brownian motions, and has no transac
tion costs, it is analyzable in this framework.
6.1 General stock model
We recall that the BlackScholes model contained a bond and a stock B
t
and S
t
with
SDEs
dB
t
= rB
t
dt,
and dS
t
= S
t
(σdW
t
+ µdt).
Here r is the constant interest rate, σ is the constant stock volatility and µ is the
constant stock drift, and we are using the SDE formulation discussed in section 4.4.
The process W is PBrownian motion.
Our most general stochastic process can have variable drift and volatility. Not
only can they vary with time, but they can depend on movements of the stock itself
(or equivalently, on movements of the Brownian motion W). We could replace the
constant σ by a function of the stock price σ(S
t
), or even a function of both the stock
price and time σ(S
t
, t). Even this is not fully general. (For instance the volatility at
time t might depend on the maximum value achieved by the stock price up to time t.)
We will replace a by a general Fprevisible process σ
t
, and the constants r and µ by
Fprevisible processes r
t
and µ
t
respectively. The new SDEs are now
dB
t
= r
t
B
t
dt,
and dS
t
= S
t
(σ
t
dW
t
+ µ
t
dt).
139
140 CHAPTER 6. BIGGER MODELS
These have solutions
B
t
= exp
__
t
0
r
s
ds
_
,
S
t
= S
0
exp
__
t
0
σ
s
dW
s
+
_
t
0
(µ
s
−
1
2
σ
2
s
)ds
_
.
[Technical note: the processes σ
t
, r
t
and µ
t
cannot be ﬁlly general, as they must
be integrable enough for these integrals to exist. Explicitly, we need that (with P
probability one), the integrals
_
T
0
σ
2
t
dt,
_
T
0
r
t
dt, and
_
T
0
µ
t
dt are ﬁnite.]
Change of measure
As before, we aim to make the discounted stock price Z
t
= B
−1
t
S
t
into a martingale.
This is achieved by adding a drift γ
t
to W. That is, if
˜
W
t
= W
t
+
_
t
0
γ
s
ds is QBrownian
motion, then Z
t
has SDE
dZ
t
= Z
t
_
σ
t
d
˜
W
t
+ (µ
t
−r
t
−σ
t
γ
t
) dt
_
.
And Z is a Qmartingale if
γ
t
=
µ
t
−r
t
σ
t
,
as was adumbrated in the market price of risk section (4.4). Now the market price
of risk depends on the time t and the sample path up to that time. It will, however,
continue to be independent of the instrument considered. It should also be checked,
in any actual case, that γ
t
satisﬁes the CMG growth condition E
P
_
exp
1
2
_
T
0
γ
2
t
dt
_
<
∞.
Under Q, Z has the SDE
dZ
t
= σ
t
Z
t
d
˜
W
t
,
so it is at least a local martingale because it is driftless. It should also be checked that
Z is a proper martingale. For instance, it is enough that E
Q
_
exp
1
2
_
T
0
σ
2
t
dt
_
is ﬁnite.
Replicating strategies
If X is the derivative to be priced, with maturity at time T, then the procedure is not
much different from the basic BlackScholes technique.
We can form a Qmartingale E
t
through the conditional expectation process of the
discounted claim, E
t
= E
Q
_
B
−1
T
XF
t
_
Then the martingale representation theorem
(section 3.5) says that the martingale E
t
is the integral
E
t
= E
0
+
_
t
0
φ
s
dZ
s
,
for some Fprevisible process φ
t
. (Note that we need σ
t
never to be zero.) Let us
take φ
t
to be our stock portfolio holding at time t. Then
dE
t
= φ
t
dZ
t
.
6.2. LOGNORMAL MODELS 141
Setting the bond portfolio holding ψ
t
to be ψ
t
= E
t
− φ
t
Z
t
, then the value of the
portfolio at time t is
V
t
= φ
t
S
t
+ ψ
t
B
t
= B
t
E
t
.
It also follows (as in chapter three) that (φ, ψ) is selfﬁnancing in that the changes in
the value V
t
are due only to changes in the assets’ prices. That is
dV
t
= φ
t
dS
t
+ ψ
t
dB
t
.
So (φ, ψ) is a selfﬁnancing strategy with initial value V
0
= E
Q
_
B
−1
T
X
_
and terminal
value V
T
= X.
Derivative pricing
Arbitrage arguments convince us that the only value for the derivative at time t is
Derivative price
V
t
= B
t
E
Q
_
B
−1
T
XF
t
_
= E
Q
_
exp
_
−
_
T
t
r
s
ds
_
X
¸
¸
¸
¸
¸
F
t
_
.
In other words, the value at time t is the suitably discounted expectation of the
derivative conditional on the history up to time t, under the measure which makes the
discounted stock process a martingale — the riskneutral measure.
There is no general expression which will provide a more explicit answer for the
option value V
t
. To make speciﬁc calculations, one needs to know the discount rate
r
t
, the volatility of the stock σ
t
— though not its drift — and the derivative itself.
Implementation
In practice, if the model is much more complex than BlackScholes, these expecta
tions cannot be performed analytically. (The lognormal cases of section 6.2 will be
notable exceptions.) Instead numerical methods must be used.
If we can approximate the price V
t
at time t, then an approximation for φ
t
or
“dV
t
/dS
t
” is the delta hedge
φ
t
≈
∆V
t
∆S
t
,
where ∆ represents the change over a small time interval (t, t + ∆t).
6.2 Lognormal models
We have already seen that the BlackScholes formula can be true, even if we are not
working with the BlackScholes model (as in section 4.1). The common feature of
models where this happens is that the asset prices are lognormally distributed under
the martingale measure Q.
142 CHAPTER 6. BIGGER MODELS
In the simple BlackScholes model, the cash bond and the stock are modeled as
B
t
= e
rt
and S
t
= S
0
exp (σW
t
+ µt) ,
where r, σ and µ are constants and W is PBrownian motion. The forward price to
purchase F at time T is
F = S
0
e
rT
.
And the value at time zero of an option to buy S
T
for a strike price of k is
V
0
= e
−rT
_
FΦ
_
log
F
k
+
1
2
σ
2
T
σ
√
T
_
−kΦ
_
log
F
k
−
1
2
σ
2
T
σ
√
T
__
.
Lognormal asset prices
When prices, under the martingale measure, are lognormal, there are great advan
tages. This holds for the BlackScholes model itself, for some currency and equity
models, and also for simple interest rate models.
Explicitly, suppose the stock S
T
and the cash bond B
T
are known to be jointly
lognormally distributed under the martingale measure Q. Let σ
2
1
T be the variance of
log S
T
, σ
2
2
T be the variance of log B
−1
T
, (σ
1
and σ
2
are term volatilities), and let ρ be
their correlation. Then the forward price for purchasing S at time T is
F =
E
Q
_
B
−1
T
S
T
_
E
Q
_
B
−1
T
_ , or equivalently F = exp (ρσ
1
σ
2
T) E
Q
(S
T
) ,
and the price of a call on S
T
struck at k is the generalized BlackScholes formula
V
0
= E
Q
_
B
−1
T
_
_
FΦ
_
log
F
k
+
1
2
σ
2
1
T
σ
1
√
T
_
−kΦ
_
log
F
k
−
1
2
σ
2
1
T
σ
1
√
T
__
.
We can see why these formulae are true. Write S
T
as
S
T
= Aexp
_
α
1
Z −
1
2
α
2
1
_
, with α
2
1
= σ
2
1
T,
where A is the constant E
Q
(S
T
) and Z is a normal N(0, 1) random variable under Q.
The discount factor is lognormal with logvariance σ
2
2
T and its correlation with the
stock logprice is ρ. Setting B to be its expectation B = E
Q
(B
−1
T
), we get
B
−1
T
= Bexp
_
α
2
(ρZ + ¯ ρW) −
1
2
α
2
2
_
, with α
2
2
= σ
2
2
T,
where ¯ ρ =
_
1 −ρ
2
and W is a normal N(0, 1) independent of Z.
The expected discounted stock price is then
E
Q
_
B
−1
T
S
T
_
= ABexp
_
1
2
(α
1
+ ρα
2
)
2
+
1
2
¯ ρ
2
α
2
2
−
1
2
α
2
1
−
1
2
α
2
2
_
= ABexp (ρα
1
α
2
) .
So the forward price for S
T
is thus F = Aexp(ρσ
1
σ
2
T). Reexpressing S
T
:
S
T
= F exp
_
α
1
Z −
1
2
α
2
1
−ρα
1
α
2
_
,
6.3. MULTIPLE STOCK MODELS 143
gives us the call value
V
0
= E
Q
_
B
−1
T
(S
T
−k)
+
_
= BE
Q
_
e
ρα
2
Z−
1
2
ρ
2
α
2
2
(S
T
−k)
+
_
,
which is also equal to
BE
Q
_
Fe
(α
1
+ρα
2
)Z−
1
2
(α
1
+ρα
2
)
2
−ke
ρα
2
Z−
1
2
ρ
2
α
2
2
; Z −z
_
,
where z is the critical value z =
_
log
F
k
−
1
2
α
2
1
−ρα
1
α
2
_
/α
1
. Using the probabilistic
result that E
_
e
yZ−
1
2
y
2
; Z −z
_
= Φ(y + z), for any constants y and z, the result
follows. [The notation E(X; A) denotes the expectation of the random variable X
over the event A, or equivalently is E(XI
A
), where I
A
is the indicator function of the
event A.]
6.3 Multiple stock models
BlackScholes assumes a single stock in the market. In many cases, this assumption
does little harm. If we write an option on, say, General Motors stock, having mod
eled its behavior adequately, we are unaffected by the movements of other securities.
However, more complex equity products, such as quantos, depend on the behavior
of at least two separate securities. Even more so in the bond market, where a swap’s
current value is affected by the movements of a large number of bonds of varying
maturities.
A good model of several securities must not only describe each one individually,
but also represent the interaction and dependency between them. For instance, our
quanto contract of section 4.5 was related to both the sterling/dollar exchange rate and
an individual UK stock. These two processes have some degree of codependence.
In particular, large movements in one may be linked with corresponding movements
in the other. Such changes would suggest that the two securities are correlated.
Stochastic processes adapted to ndimensional Brownian motion
A stochastic process X is a continuous process (X
t
: t ≥ 0) such that X
t
can be
written as
X
t
= X
0
+
n
i=1
_
t
0
σ
i
(s)dW
i
s
+
_
t
0
µ
s
ds,
where σ
1
, . . . , σ
n
and µ are random Fprevisible processes such that the integral
_
t
0
_
i
σ
2
i
(s) +µ
s

_
ds is ﬁnite for all times t (with probability 1). The differential
form of this equation can be written
dX
t
=
n
i=1
σ
i
(t) dW
i
t
+ µ
t
dt.
Multiple stocks can be driven by multiple Brownian motions. Instead of just one
PBrownian motion, we will have, in the nfactor case, n independent Brownian
144 CHAPTER 6. BIGGER MODELS
motions W
1
t
, . . . , W
n
t
. That means that each W
i
t
behaves as a Brownian motion, and
the behavior of any one of them is completely uninﬂuenced by the movements of
the others. Their ﬁltration F
t
is now the total of all the histories of the n Brownian
motions. In other words, F
T
is the history of the ndimensional vector (W
1
t
, . . . , W
n
t
)
up to time T. This leads to an enhanced deﬁnition of a stochastic process (see box).
The drift term is unchanged from the original (onefactor) deﬁnition, but there is
now a volatility process σ
i
(t) for each factor. We must remember that in a multi
factor setting volatility is no longer a scalar, but strictly is now a vector. The total
volatility of the process X is
_
σ
2
1
(t) +· · · + σ
2
n
(t). In other words, the variance of
dX
t
is
i
σ
2
i
(t)dt, made up of the contribution σ
2
i
(t)dt from each Brownian motion
component W
i
, the variances adding because the Brownian motion components are
independent.
There is also an nfactor version of It ˆ o’s formula and the product rule.
Itˆ o’s formula (nfactor)
If X is a stochastic process, satisfying dX
t
=
i
σ
i
(t) dW
i
t
+ µ
t
dt, and f is a
deterministic twice continuously differentiable function, then Y
t
:= f(X
t
) is also a
stochastic process with stochastic increment
dY
t
=
n
i=1
_
σ
i
(t) f
(X
t
)
_
dW
i
t
+
_
µ
t
f
(X
t
) +
1
2
n
i=1
σ
2
i
(t) f
(X
t
)
_
dt.
Again this is an analogue of the onefactor It ˆ o formula, with the replication of the
volatility terms for each additional Brownian factor.
Product rule (nfactor)
If X is a stochastic process satisfying dX
t
=
i
σ
i
(t) dW
i
t
+ µ
t
dt, and Y is a
stochastic process satisfying dY
t
=
i
ρ
i
(t) dW
i
t
+ ν
t
dt, then X
t
Y
t
is a stochastic
process satisfying
d (X
t
Y
t
) = X
t
dY
t
+ Y
t
dX
t
+
_
n
i=1
σ
i
(t) ρ
i
(t)
_
dt.
This new version uniﬁes the two apparently different cases of the product rule we
encountered in section 3.3. If X
t
and Y
t
are both adapted to the same Brownian mo
tion W
t
, then this rule agrees with the ﬁrst case. If however X
t
and Y
t
are adapted to
two independent Brownian motions, say W
1
t
and W
2
t
, then X
t
will have zero volatility
with respect to W
2
, that is σ
2
(t) = 0, and similarly Y
t
will have zero volatility with
respect to W
1
, ρ
1
(t) = 0. Thus the term
σ
i
(t)ρ
i
(t) in the nfactor product rule will
be identically zero, agreeing with the second case in section 3.3.
The CameronMartinGirsanov theoremcontinues to hold where W is ndimensional
Brownian motion and the drift y is an nvector process for which E
P
exp
_
1
2
_
T
0
γ
t

2
dt
_
6.3. MULTIPLE STOCK MODELS 145
is ﬁnite.
CameronMartinGirsanov theorem (nfactor)
Let W = (W
1
, . . . , W
n
) be ndimensional PBrownian motion. Suppose that
γ
t
= (γ
1
t
, . . . , γ
n
t
) is an Fprevisible nvector process which satisﬁes the growth
condition E
P
exp
_
1
2
_
T
0
γ
t

2
dt
_
< ∞, and we set
˜
W
i
t
= W
i
t
+
_
t
0
γ
i
s
ds. Then there
is a new measure Q, equivalent to P up to time T, such that
˜
W :=
_
˜
W
1
, . . . ,
˜
W
n
_
is ndimensional QBrownian motion up to time T.
The RadonNikodym derivative of Q by P is
dQ
dP
= exp
_
−
n
i=1
_
T
0
γ
i
t
dW
i
t
−
1
2
_
T
0
γ
t

2
dt
_
.
There is also a converse to this theorem, exactly analogous to the one factor con
verse.
Finally, we recall from section 5.5 that there is an nfactor martingale representa
tion theorem. With W as ndimensional QBrownian motion, M as an ndimensional
Qmartingale with nonsingular volatility matrix, and N any other onedimensional
Qmartingale, then there is an Fprevisible nvector process φ
t
= (φ
1
t
, . . . , φ
n
t
) such
that
N
t
= N
0
+
n
j=1
_
t
0
φ
j
s
dM
j
s
.
The general nfactor model
We will see later that it is important that we have essentially as many basic securities
(excluding the cash bond) as there are Brownian factors. Generally speaking, if there
are more securities than factors there might be arbitrage, and if there are fewer we
will not be able to hedge. The situation is not quite as simple as that (the bond market,
for instance, has an unlimited number of different maturity bonds), but we shall start
with the canonical case.
Our model then, will contain a cash bond B
t
as usual, and n different market
securities S
1
t
, . . . , S
n
t
. Their SDEs are
dB
t
= r
t
B
t
dt,
dS
i
t
= S
i
t
_
_
n
j=1
σ
ij
(t) dW
i
t
+ µ
i
t
dt
_
_
, i = 1, . . . , n.
Here r
t
is the instantaneous shortrate process, µ
i
t
is the drift of the ith security, and
(σ
ij
)
n
j=1
is its volatility vector. As each security has a volatility vector, the collection
of n such vectors forms a volatility matrix Σ
t
= (σ
ij
(t))
n
i,j=1
of processes. In integral
146 CHAPTER 6. BIGGER MODELS
form, these securities are
B
t
= exp
__
t
0
r
s
ds
_
,
S
i
t
= S
i
0
exp
_
_
_
n
j=1
_
t
0
σ
ij
(s)dW
j
s
+
_
t
0
_
_
µ
i
s
−
1
2
n
j=1
σ
2
ij
(s)
_
_
ds
_
_
_
Change of measure
We now want to ﬁnd a new measure Q, under which all the discounted stock prices
are Qmartingales simultaneously.
Suppose we add a drift γ
t
= (γ
1
t
, . . . , γ
n
t
) to W
t
, so that
˜
W
i
t
= W
i
t
+
_
t
0
γ
i
s
ds
is QBrownian motion, by the nfactor CMG theorem. Then the discounted stock
price Z
i
t
= B
−1
t
S
i
t
has SDE
dZ
i
t
= Z
i
t
_
_
n
j=1
σ
ij
(t) d
˜
W
j
t
+
_
_
µ
i
t
−r
t
−
n
j=1
σ
ij
(t) γ
j
t
_
_
dt
_
_
.
To make the drift term vanish for each i, we must have that
n
j=1
σ
ij
(t) γ
i
t
= µ
i
t
−r
t
, for all t, i = 1, . . . , n.
In terms of vectors and matrices, this can be reexpressed as
Σ
t
γ
t
= µ
t
−r
t
1,
where Σ
t
is the matrix
_
σ
ij
(t)
_
and 1 is the constant vector (1, 1, . . . , 1). This vector
equation may or may not have a solution γ
t
for any particular t. Whether it does or not
depends on the actual values of Σ
t
, µ
t
and r
t
. If, though, the matrix Σ
t
is invertible,
then a unique such γ
t
must exist and be equal to
γ
t
= Σ
−1
t
(µ
t
−r
t
1) .
The onefactor market price of risk formula γ
t
= σ
−1
t
(µ
t
− r
t
) is now just a spe
cial case. This means that if Σ
t
is invertible for every t and γ
t
satisﬁes the CMG
condition E
P
exp
_
1
2
_
T
0
γ
t

2
dt
_
< ∞, then there is a measure Q which makes the
discounted stock prices into Qmartingales. (Or at least into Qlocal martingales. We
also need the integral condition that for each i, E
Q
_
exp
1
2
n
j
_
T
0
σ
2
ij
(t) dt
_
< ∞, for
Z
i
to be a proper Qmartingale.)
6.4. NUMERAIRES 147
Replicating strategies
Let X be a derivative maturing at time T, and let E
t
be the Qmartingale E
t
=
E
Q
_
B
−1
T
XF
t
_
If the matrix Σ
t
is always invertible, then the nfactor martingale rep
resentation theorem gives us a volatility vector process φ
t
= (φ
1
t
, . . . , φ
n
t
) such that
E
t
= E
0
+
n
j=1
_
t
0
φ
j
s
dZ
j
s
.
The invertibility of Σ
t
is essential at this stage. Our hedging strategy will be (φ
1
t
, . . . , φ
n
t
, ψ
t
)
where φ
i
t
is the holding of security i at time t and ψ
t
is the bond holding. As usual,
the bond holding ψ is
ψ
t
= E
t
−
n
j=1
φ
j
t
Z
j
t
,
so that the value of the portfolio is V
t
= B
t
E
t
. The portfolio is selfﬁnancing in that
dV
t
=
n
j=1
φ
j
t
dS
j
t
+ ψ
t
dB
t
.
'
&
$
%
Derivative pricing
The value of the derivative at time t is
V
t
= B
t
E
Q
_
B
−1
T
XF
t
_
= E
Q
_
exp
_
−
_
T
t
r
s
ds
_
X
¸
¸
¸
¸
¸
F
t
_
.
6.4 Numeraires
Although the numeraire is usually chosen to be a cash bond, it needn’t be. In fact,
not only can the numeraire have volatility, it can be any of the tradable instruments
available. We have seen in the foreign exchange context that there can be a choice
of which currency’s cash bond to use. But no matter which numeraire is chosen, the
price of the derivative will always be the same. It is because the choice of numeraire
doesn’t matter, that we usually pick the stolid cash bond.
When we proved the selfﬁnancing condition in chapter three, we assumed that
the numeraire had no volatility. This is not actually necessary. But we do have to
check that the selfﬁnancing equations will still work. We want to show that
Selfﬁnancing strategies
A portfolio strategy (φ
t
, ψ
t
) of holdings in a stock S
t
and a possibly volatile cash
bond B
t
has value V
t
= φ
t
S
t
+ψ
t
B
t
and discounted value E
t
= φ
t
Z
t
+ψ
t
, where Z
is the discounted stock process Z
t
= B
−1
t
S
t
. Then the strategy is selfﬁnancing if
either
dV
t
= φ
t
dS
t
+ ψ
t
dB
t
, or equivalently dE
t
= φ
t
dZ
t
.
148 CHAPTER 6. BIGGER MODELS
Recall the onefactor product rule
d (XY )
t
= X
t
dY
t
+ Y
t
dX
t
+ σ
t
ρ
t
dt,
where X and Y are stochastic processes with stochastic differentials
dX
t
= σ
t
dW
t
+ µ
t
dt,
dY
t
= ρ
t
dW
t
+ ν
t
dt.
Suppose we have a strategy (φ, ψ), with discounted value E
t
satisfying dE
t
=
φ
t
dZ
t
. We want to show that (φ, ψ) is selfﬁnancing. We do this with two applications
of the product rule. Firstly
dV
t
= d (B
t
E
t
) = B
t
dE
t
+ E
t
dB
t
+ σ
t
(φ
t
ρ
t
) dt,
where σ
t
is the volatility of B
t
and ρ
t
is the volatility of Z
t
(and hence φ
t
ρ
t
is the
volatility of E
t
). We can use the substitutions dE
t
= φ
t
dZ
t
and E
t
= φ
t
Z
t
+ ψ
t
to
rearrange the above expression into
dV
t
= φ
t
(B
t
dZ
t
+ Z
t
dB
t
+ σ
t
ρ
t
dt) + ψ
t
dB
t
.
The second use of the product rule says that the term in brackets above is equal to
d(BZ)
t
= dS
t
. The resulting equation is the selfﬁnancing equation.
This also holds for nfactor models with multiple stocks.
Changing numeraires
Suppose we have a number of securities including some stocks S
1
t
, . . . , S
n
t
and two
others B
t
and C
t
either of which might be a numeraire. If we choose B
t
to be our
numeraire, we need to ﬁnd a measure Q (equivalent to the original measure) under
which
B
−1
t
S
i
t
(i = 1, . . . , n) and B
−1
t
C
t
are Qmartingales. Then the value at time t of a derivative payoff X at time T is
V
t
= B
t
E
Q
_
B
−1
T
XF
t
_
.
Suppose however that we choose C
t
to be our numeraire instead. Then we would
have a different measure Q
C
under which
C
−1
t
S
i
t
(i = 1, . . . , n) and C
−1
t
B
t
are Q
C
martingales. We can actually ﬁnd out what Q
C
is, or at least what its Radon
Nikodym derivative with respect to Q is. We recall RadonNikodym fact (ii) from
section 3.4, that for any process X
t
,
ζ
s
E
Q
C (X
t
F
s
) = E
Q
(ζ
t
X
t
F
s
) ,
where ζ
t
is the change of measure process ζ
t
= E
Q
_
dQ
C
dQ
¸
¸
¸ F
t
_
. It follows from this
that if X
t
happens to be a Q
C
martingale, then
ζ
s
X
s
= E
Q
(ζ
t
X
t
F
s
) ,
6.4. NUMERAIRES 149
and so ζ
t
X
t
is a Qmartingale.
The canonical Q
C
martingales (including the constant martingale with value 1) are
1, C
−1
t
B
t
, C
−1
t
S
1
t
, . . . , C
−1
t
S
n
t
and similarly the Qmartingales are B
−1
t
C
t
, 1, B
−1
t
S
1
t
, . . . , B
−1
t
S
n
t
.
Each corresponding pair has a common ratio of ζ
t
= B
−1
t
C
t
. Thus the Radon
Nikodym derivative of Q
C
with respect to Q is the ratio of the numeraire C to the
numeraire B,
dQ
C
dQ
=
C
T
B
T
.
The price of a payoff X maturing at T under the Q
C
measure is
V
C
t
= C
t
E
Q
C
_
C
−1
T
XF
t
_
.
Using again the RadonNikodym result that E
Q
C (XF
t
) = ζ
−1
t
E
Q
(ζ
T
XF
t
), then
V
C
t
= ζ
−1
t
C
t
E
Q
_
ζ
T
C
−1
T
XF
t
_
= B
t
E
Q
_
B
−1
T
XF
t
_
.
This is exactly the same as the price V
t
under Q, so the two agree, just as in the
foreign exchange section (4.1), where the dollar and sterling investors agreed on all
derivative prices.
Example — forward measures in the interestrate market
In interestrate models, it is often popular to use a bond maturing at date T (the
Tbond with price P(t, T)) as the numeraire. The martingale measure for this nu
meraire is called the Tforward measure P
T
and makes the forward rate f(t, T) a
P
T
martingale, as well as the δperiod LIBOR rate for borrowing up till time T.
The new numeraire is the Tbond normalized to have unit value at time zero. If
we call this numeraire C
t
, then C
t
= P(t, T)/P(0, T). The forward measure P
T
thus
has RadonNikodym derivative with respect to Q of
dP
T
dQ
−
C
T
B
T
−
1
P (0, T) B
T
.
The associated Qmartingale is
ζ
t
= E
Q
_
dP
T
dQ
¸
¸
¸
¸
F
t
_
=
C
t
B
t
=
P (t, T)
P (0, T) B
t
.
Now the forward price set at time t for purchasing X at date T is its current value V
t
scaled up by the return on a Tbond, namely F
t
= P
−1
(t, T) B
t
E
Q
_
B
−1
T
XF
t
_
. Once
more, by property (ii) of the RadonNikodym derivative, F
t
equals
F
t
= E
P
T
(XF
t
) ,
so is itself a P
T
martingale. Calculating the forward price for X is now only a matter
of taking its expectation under the forward measure.
From the SDE for P(t, T), we ﬁnd that ζ
t
satisﬁes
dζ
t
= ζ
t
n
i=1
Σ
i
(t, T) dW
i
(t),
150 CHAPTER 6. BIGGER MODELS
where W is ndimensional QBrownian motion, and Σ
i
(t, T) is the component of the
volatility of P(t, T) with respect to W
i
(t). By the converse of the CMG theorem,
we see that
˜
W
i
(t) = W
i
(t) −
_
t
0
Σ
i
(s, T) ds
is P
T
Brownian motion.
This gives an alternative expression for pricing interestrate derivatives. If X is a
payoff at date T, then its value at time t is
V
t
= B
t
E
Q
_
B
−1
T
XF
t
_
= P (t, T) E
P
T
(XF
t
) .
So the value of X at time t is just the P
T
expectation of X up to time t (the forward
price of X) discounted by the (Tbond) time value of money up to date T.
Also the forward rates f(t, T) are the forward rates for r
T
, so that f(t, T) is a
P
T
martingale with
f (t, T) = E
P
T
(r
T
F
t
) ,
and d
t
f (t, T) =
n
i=1
σ
i
(t, T) d
˜
W
i
(t) .
Another forward measure martingale is the δperiod LIBOR rate
L
t
=
1
δ
_
P (t, T −δ)
P (t, T)
−1
_
.
See chapter ﬁve (section 5.7) for more details.
6.5 Foreign currency interestrate models
We have looked at foreign exchange (section 4.1). We have looked at the interest rate
market (chapter ﬁve). But we have not yet studied an interest rate market of another
currency. Now we will.
For deﬁniteness, we will imagine ourselves to be a dollar investor operating in
both the dollar and sterling interestrate markets. Our variables will be:
As in the HJMmodel, we will work in an nfactor model driven by the independent
Brownian motions W
1
t
, . . . , W
n
t
. Of course n might be one, but it needn’t be, in which
case, the volatilities σ, τ and ρ are nvectors σ
i
(t, T), τ
i
(t, T) and ρ
i
(t) (i = 1, . . . , n).
What we have here are two separate interestrate markets (the dollar denominated
and the sterling denominated), plus a currency market linking them. The multifactor
model approach is needed to reﬂect varying degrees of correlation between various
securities in the three markets.
6.5. FOREIGN CURRENCY INTERESTRATE MODELS 151
Table 6.1: Notation
P(t, T) : the dollar zerocoupon bond market prices
f(t, T) : the forward rate of dollar borrowing at date T (is −
∂
∂T
log P (t, T))
σ(t, T) : the volatility of f(t, T)
α(t, T) : the drift of f(t, T)
r
t
: the dollar short rate (equal to f(t, t))
B
t
: the dollar cash bond (equal to exp
_
t
0
r
s
ds)
Q(t, T) : the sterling zerocoupon bond market prices
g(t, T) : the forward rate of sterling borrowing at date T (is −
∂
∂T
log Q(t, T))
τ(t, T) : the volatility of g(t, T)
β(t, T) : the drift of g(t, T)
u
t
: the sterling short rate (equal to g(t, t))
D
t
: the sterling cash bond (equal to exp
_
t
0
u
s
ds)
C
t
: the exchange rate value in dollars of one pound
ρ
t
: the logvolatility of the exchange rate
λ
t
: the drift coefﬁcient of the exchange rate (the drift of dC
t
/C
t
).
The differentials of these processes are
d
t
f (t, T) =
n
i=1
σ
i
(t, T) dW
i
t
+ α(t, T) dt,
d
t
g (t, T) =
n
i=1
τ
i
(t, T) dW
i
t
+ β (t, T) dt,
dC
t
= C
t
_
n
i=1
ρ
i
(t) dW
i
t
+ λ
t
dt
_
.
Apart from the dollar cash bond B
t
, the dollar tradable securities in this market con
sist of the dollarbonds P(t, T); the dollar worth of the sterling bonds C
t
Q(t, T); and
the dollar worth of the sterling cash bond C
t
D
t
. Let us ﬁx T, and let the dollar dis
counted value of these three securities be X, Y and Z respectively, where
X
t
= B
−1
t
P (t, T) ,
Y
t
= B
−1
t
C
t
Q(t, T) ,
Z
t
= B
−1
t
C
t
D
t
.
It will simplify later expressions to introduce the notation Σ
i
, T
i
and
˜
T
i
, where
Σ
i
(t, T) = −
_
T
t
σ
i
(t, u) du,
T
i
(t, T) = −
_
T
t
τ
i
(t, u) du,
˜
T
i
(t, T) = T
i
(t, T) + ρ
i
(t) .
Then Σ
i
(t, T) is the W
i
volatility term of P(t, T), T
i
(t, T) is the same for Q(t, T), and
˜
T
i
(t, T) is the same for C
t
Q(t, T).
152 CHAPTER 6. BIGGER MODELS
Our plan, much as ever, is to follow the three steps to replication. The ﬁrst thing
to do is to ﬁnd a change of measure under which X
t
, Y
t
, and Z
t
are all martingales.
For any previsible nvector γ = (γ
i
(t))
n
i=1
, there is a new measure Q and a Q
Brownian motion
˜
W = (
˜
W
1
t
, . . . ,
˜
W
n
t
), where
˜
W
i
t
= W
i
t
+
_
t
0
γ
i
(s)ds. Then the SDEs
of X, Y and Z with respect to Q are
dX
t
= X
t
_
n
i=1
Σ
i
(t, T) d
˜
W
i
t
+
_
_
T
t
(ξ (t, u) −α(t, u)) du
_
dt
_
dY
t
= Y
t
_
n
i=1
˜
T
i
(t, T) d
˜
W
i
t
+
_
ν
t
+
_
T
t
(η (t, u) −β (t, u)) du
_
dt
_
dZ
t
= Z
t
_
n
i=1
ρ
i
(t) d
˜
W
i
t
+ ν
t
dt
_
,
where ξ(t, T), η(t, T) and ν
t
are deﬁned to be
ξ (t, T) =
n
i=1
σ
i
(t, u) (γ
i
(t) −Σ
i
(t, u)) ,
η (t, T) =
n
i=1
τ
i
(t, u)
_
γ
i
(t) −
˜
T
i
(t, u)
_
,
ν
t
= λ
t
−r
t
+ u
t
−
i
ρ
i
(t) γ
i
(t) .
Then there will be a martingale measure only if there is some choice of γ which
makes all of X, Y and Z driftless. This happens if
α(t, T) =
n
i=1
σ
i
(t, T) (γ
i
(t) −Σ
i
(t, T)) ,
β (t, T) =
n
i=1
τ
i
(t, T)
_
γ
i
(t) −
˜
T
i
(t, T)
_
,
λ
t
= r
t
−u
t
+
n
i=1
ρ
i
(t) γ
i
(t).
Then under this Q measure
d
t
P (t, T) = P (t, T)
_
n
i=1
Σ
i
(t, T) d
˜
W
i
t
+ r
t
dt
_
,
d
t
Q(t, T) = Q(t, T)
_
n
i=1
T
i
(t, T) d
˜
W
i
t
+
_
u
t
−
n
i=1
ρ
i
(t) T
i
(t, T)
_
dt
_
,
dC
t
= C
t
_
n
i=1
ρ
i
(t) d
˜
W
i
t
+ (r
t
−u
t
) dt
_
.
As long as this measure Q is unique, we will be able to hedge. (And uniqueness
will follow if the volatility vectors of any n of the dollar tradable securities make an
invertible matrix.) A derivative X paid in dollars at date T will have value at time t
V
t
= B
t
E
Q
_
B
−1
T
XF
t
_
.
6.6. ARBITRAGEFREE COMPLETE MODELS 153
The sterling investor
The sterling investor is on the other side of the mirror. He works with a different
martingale measure Q
£
. This reﬂects that his numeraire is the sterling cash bond D
t
rather than the dollar cash bond. The RadonNikodym derivative of Q
£
with respect
to Q will be the ratio of the dollar worth of the sterling bond to the dollar numeraire.
(Normalizing D
0
= 1/C
0
for convenience.) That is
E
Q
_
dQ
£
dQ
¸
¸
¸
¸
F
t
_
=
C
t
D
t
B
t
= Z
t
.
As Z
t
has the SDE dZ
t
= Z
t
i
ρ
i
(t) d
˜
W
i
t
the difference in drifts between the Q
£

Brownian motion
˜
W
£
and the QBrownian motion
˜
W is just ρ. That is
˜
W
£
i
(t) =
˜
W
i
t
−
_
t
0
ρ
i
(s)ds.
To the sterling investor, the sterling bonds have SDE
d
t
Q(t, T) = Q(t, T)
_
n
i=1
T
i
(t, T) d
˜
W
£
i
(t) + u
t
dt
_
,
which is exactly the form that HJM leads us to expect.
As explained in section 6.4, the sterling investor will agree with the dollar investor
on prices of future payoffs.
6.6 Arbitragefree complete models
Time and again we have seen the same basic techniques used to price and hedge
derivatives. Firstly, the CMG theorem is used to make the discounted price pro
cesses into martingales under a new measure. Then the martingale representation
theorem gives a hedge for the derivative. The repeated recurrence of this program
suggests that there might be a more general result underpinning it. And there is.
Before stating this canonical theorem, it is worth carefully laying out some con
cepts we have already brushed up against.
• arbitragefree. A market is arbitragefree if there is no way of making riskless
proﬁts. An arbitrage opportunity would be a (selfﬁnancing) trading strategy
which started with zero value and terminated at some deﬁnite date T with a pos
itive value. A market is arbitragefree if there are absolutely no such arbitrage
opportunities.
• complete. A market is said to be complete if any possible derivative claim can
be hedged by trading with a selfﬁnancing portfolio of securities.
• equivalent martingale measure (EMM). Suppose we have a market of se
curities and a numeraire cash bond under a measure P. An EMM is a mea
sure Q equivalent to P, under which the bonddiscounted securities are all Q
martingales. This is just a more precise name for what we call the martingale
measure.
154 CHAPTER 6. BIGGER MODELS
Already we have examples of the binomial trees and the continuoustime Black
Scholes model. Both of these are complete markets with an EMM. We have not
found an arbitrage opportunity, but neither are we sure that one might not exist.
In both the binomial tree and BlackScholes models we found there was one and
only one EMM, and we were able to hedge claims. Even more so in the multiple
stock models (section 6.3). There we could ﬁnd a market price of risk γ
t
but it (and
so Q too) was only unique if the volatility matrix Σ
t
was invertible. And it was ex
actly that invertibility which lets us hedge.
Arbitragefree and completeness theorem (Harrison and Pliska)
Suppose we have a market of securities and a numeraire bond. Then
(1) the market is arbitragefree if and only if there is at least one EMMQ; and
(2) in which case, the market is complete if and only if there is exactly one such
EMMQ and no other.
This simple yet powerful theorem makes sense of our experience.
In the HJM bondmarket model, these conditions were also visible. The model
demands that the forward rate drift α(t, T) satisﬁed
α(t, T) =
n
i=1
σ
i
(t, T) (γ
i
(t) −Σ
i
(t, T)),
for some previsible processes γ
i
(t). This ensures that there is an EMM Q, and γ
is the market price of risk. We now see that this is to make sure that the model is
arbitragefree.
The other key HJM condition is that the volatility matrix
_
Σ
i
(t, T
j
)
_
n
i,j=1
is nonsingular for all sequences of dates T
1
< · · · < T
n
, and for all t less than T
1
,
which means there is only one viable price of risk in the market. This is sufﬁcient (but
actually slightly more than necessary) for the EMM to be unique, and consequently
for the market to be complete.
It is worth getting a feel of why this theorem works. Although the technical details
and exact deﬁnitions are passed over, the structure of the following can be proved
rigorously.
Martingales mean no arbitrage
A martingale is really the essence of a lack of arbitrage. The governing rule for a
Qmartingale M
t
is that
E
Q
(M
t
F
s
) = M
s
.
In other words, its future expectation, given the history up to time s, is just its current
value at time s. The martingale is not ‘expected’ to be either higher or lower than its
6.6. ARBITRAGEFREE COMPLETE MODELS 155
present value. An arbitrage opportunity, on the other hand, is a oneway bet which is
certain to end up higher than it started.
Suppose we have a potential arbitrage opportunity contained in the selfﬁnancing
portfolio strategy (φ, ψ). (Assuming for simplicity a two security market of stock S
t
and bond B
t
.) Then its value at time t is
V
t
= φ
t
S
t
+ ψ
t
B
t
,
and it satisﬁes the selfﬁnancing equation
dV
t
= φ
t
dS
t
+ ψ
t
dB
t
.
We can calculate the discounted value of the portfolio E
t
= B
−1
t
V
t
, and then
dE
t
= φ
t
dZ
t
,
where Z
t
is the discounted stock price B
−1
t
S
t
which is a Qmartingale.
Suppose now that the strategy does start with zero value (V
0
= 0) and ﬁnishes
with a nonnegative payoff (V
T
≥ 0). Can this really be an arbitrage opportunity?
Crucially, E
t
is a Qmartingale because Z
t
is. And so
E
Q
(E
T
) = E
Q
(E
T
F
0
) = E
0
= V
0
= 0.
But V
T
≥ 0 and (because B
−1
T
> 0) so is E
T
≥ 0. But the Qexpectation of E
T
is zero,
so the only possible value that E
T
can take is zero too.
From which it is clear that V
T
is zero as well. Any strategy can make no more
than nothing from nothing. A martingale is essentially a ‘fair game’ and any strategy
which involves only playing fair games cannot guarantee a proﬁt.
Or in our language, if an EMM exists, there are no arbitrage opportunities.
Hedging means unique prices
If we can hedge, then there can only be at most one EMM.
To see this, suppose that we could hedge, but that there are two different EMMs
Q and Q
.
For any event A in the history F
T
, the digitallike claim which pays off the cash
bond value at time T if A has happened has payoff X = B
T
I
A
. (The indicator
function I
A
takes the value 1 if the event A happens, and zero otherwise.) This is
a valid derivative, so it must be hedgeable. (We assumed that we could hedge all
claims.) So there must be a selfﬁnancing portfolio (φ, ψ) which hedges X, with
value
V
t
= φ
t
S
t
+ ψ
t
B
t
.
As usual the discounted claim E
t
= B
−1
t
V
t
satisﬁes
dE
t
= φ
t
dZ
t
,
156 CHAPTER 6. BIGGER MODELS
where Z
t
is the discounted stock price B
−1
t
S
t
. Now Z
t
is both a Qmartingale and a
Q
martingale as both Q and Q
are EMMs. So also must E
t
be. And from that, we
see
E
0
= E
Q
(E
T
) = E
Q
(E
T
) .
But E
T
is just the indicator function of the event A, I
A
, and so E
0
= Q(A) = Q
(A).
The two measures Q and Q
which were trying to be different actually give the same
likelihood for the event A. As A was completely general, the two measures agree
completely, and thus Q = Q
. If any two EMMs are identical, then there can only
really be one EMM.
Harrison and Pliska
We have only proved each result in one direction. We showed that if there was an
EMM there was no arbitrage, but did not show that if there is no arbitrage then there
actually is an EMM. Also we proved that hedging can only happen with a unique
EMM, but not that the uniqueness of the EMM forced hedging to be possible.
The full and rigorous proofs of all these results in the discretetime case are in
the paper ‘Martingales and stochastic integrals in the theory of continuous trading’
by Michael Harrison and Stanley Pliska, in Stochastic Processes and their Applica
tions (see appendix A for more details). For the continuous case and more advanced
models, there has been other work, notably by Delbaen and Schachermayer. But the
increasing technicality of this should not stand in the way of an appreciation of the
remarkable insight of Harrison and Pliska.
Appendix A
Further reading
The longer a list of books is, the fewer will actually be referred to. The lists below
have been kept short, in the hope that in this case less choice is more.
Probability and stochastic calculus books
• A ﬁrst course in probability, Sheldon Ross, Macmillan (4th edition 1994, 420
pages)
• Probability and random processes, Geoffrey Grimmett and David Stirzaker, Ox
ford University Press (2nd edition 1992, 540 pages)
• Probability with martingales, David Williams, Cambridge University Press (1991,
250 pages)
• Continuous martingales and Brownian motion, Daniel Revuz and Mark Yor,
Springer (2nd edition 1994, 550 pages)
• Diffusions, Markov processes, and martingales: vol. 2 Itˆ o calculus, Chris Rogers
and David Williams, Wiley (1987, 475 pages)
These books are arranged in increasing degrees of technicality and depth (with the
last two being at an equivalent level) and contain the probabilistic material used in
chapters one, two and three. Ross is an introduction to the basic (static) probabilistic
ideas of events, likelihood, distribution and expectation. Grimmett and Stirzaker con
tain that material in their ﬁrst half, as well as the development of random processes
including some basic material on martingales and Brownian motion.
Probability with martingales not only lays the groundwork for integration, (condi
tional) expectation and measures, but also is an excellent introduction to martingales
themselves. There is also a chapter containing a simple representation theorem and a
discretetime version of BlackScholes.
Both Revuz and Yor, and Rogers and Williams provide a detailed technical cov
erage of stochastic calculus. They both contain all our tools; stochastic differentials,
It ˆ o’s formula, CameronMartinGirsanov change of measure, and the representation
theorem. Although dense with material, a reader with background knowledge will
ﬁnd them invaluable and deﬁnitive on questions of stochastic analysis.
157
158 APPENDIX A. FURTHER READING
Financial books
• Options,futures, and other derivative securities, John Hull, PrenticeHall (2nd
edition 1993, 490 pages)
• Dynamic asset pricing theory, Darrell Dufﬁe, Princeton University Press (1992,
300 pages)
• Option pricing: mathematical models and computation, Paul Wilmott, Jeff Dewynne
and Sam Howison, Oxford Financial Press (1993, 450 pages)
Hull is a popular book with practitioners, laying out the various realworld options
contracts and markets before starting his analysis. A number of models are discussed,
and numerical procedures for implementation are also included. The chapterby
chapter bibliographies are another useful feature.
Dufﬁe is a much more mathematically rigorous text, but still accessible. He con
tains sections on equilibrium pricing and optimal portfolio selection as well as a
treatment of continuoustime arbitragefree pricing along the same lines as this book.
For readers with mathematical backgrounds, it is a good read.
Oxford Financial Press’s volume comes at the subject purely from a differential
equation framework without using stochastic techniques. Eventually, many pricing
problems become differential equation problems, but unless a reader has experience
in this area, it is not necessarily the best place to start from.
Chapter four: pricing market securities
Some notable journal papers include:
• The pricing of options and corporate liabilities, F Black and M Scholes, Journal
of Political Economy, 81 (1973), 637–654.
• Theory of rational option pricing, R C Merton, Bell Journal of Economics and
Management Science, 4 (1973), 141–183.
• Foreign currency option values, M B Garman and S W Kohlhagen, Journal of
International Money and Finance, 2 (1983), 231–237.
• Options markets, J C Cox and M Rubinstein, PrenticeHall (1985, 500 pages).
• Two into one, M Rubinstein, RISK, (May 1991), p. 49.
The BlackScholes paper is now of historical interest, but it is still fascinating
to see how the subject began, though the paper should be read for its insights, not
the technical detail. At the time they were as concerned with pricing the stock of
companies with outstanding liabilities (such as corporate bonds or warrants) as they
were about options and derivatives.
Merton provides a more rigorous treatment, contemporaneously with BlackScholes,
and makes extension to dividendpaying stocks and a barrier option. Garman and
159
Kohlhagen described foreign exchange options, whilst Cox and Rubinstein contain
some exotic option formulas, amongst much else. The Rubinstein paper from RISK
is concerned with quantos and crosscurrency options.
Chapter ﬁve: interest rates
In the interestrate setting, HeathJarrowMorton is as seminal as Black Scholes. By
focusing on forward rates and especially by giving a careful stochastic treatment,
they produced the most general (ﬁnite) Brownian interestrate model possible. Other
models may claim differently, but they are just HJM with different notation. The
paper itself repays reading and rereading.
• Bond pricing and the term structure of interest rates: a new methodology for
contingent claims valuation, David Heath, Robert Jarrow and Andrew Morton,
Econometrica, 60 (1992), 77–105.
In addition to the HJM paper, notable papers on the various interestrate market
models include
• Term structure movements and pricing interest rate contingent claims, T S Y Ho
and SB Lee, Journal of Finance, 41 (1986), 101 1–1029.
• An equilibrium characterization of the term structure, O A Vasicek, Journal of
Finance, 5 (1977), 177–188.
• Pricing interest rate derivative securities, J Hull and A White, The Review of
Financial Studies, 3 (1990), 573–592.
• A theory of the term structure of interest rates, J C Cox, J E Ingersoll and S A
Ross, Econometrica, 53 (1985), 385–407.
• Bond and option pricing when short rates are lognormal, F Black and P Karasin
ski, Financial Analysts Journal, (JulyAugust 1991), 52–59.
• The market model of interest rate dynamics, A Brace, D Gatarek and M Musiela,
UNSW Preprint, Department of Statistics S95–2.
• Which model for the termstructure of interest rates should one use? L C G
Rogers, in Mathematical Finance (ed. M H A Davis, D Dufﬁe, et al.), IMA
Volume 65, SpringerVerlag, 93–116.
The last of these is a review paper of models and their properties, whilst the others
describe separately all the major models considered in the chapter.
Chapter six: bigger models
• Martingales and stochastic integrals in the theory of continuous trading, Michael
Harrison and Stanley Pliska, Stochastic Processes and their Applications, 11
(1981), 215–260.
160 APPENDIX A. FURTHER READING
• The fundamental theorem of asset pricing, F Delbaen and W Schachermayer,
Mathematische Annalen, 300 (1994), 463–520.
• The valuation of options for alternative stochastic processes, J C Cox and S A
Ross, Journal of Financial Economics, 3 (1976), 145–166.
Harrison and Pliska made the next step forward by linking, in a general frame
work, the absence of arbitrage to the existence of a martingale measure, and showing
that the ability to hedge depended on there only being one such measure. That this
idea still underpins much of ﬁnancial mathematics today is a demonstration of the
importance of the paper.
Delbaen and Schachermayer go over similar ground but in a much more technical
way to deal with the particular problems of continuoustime processes, including
discontinuous processes. Cox and Ross cover option pricing for models more general
than BlackScholes, including those paying dividends.
Appendix B
Notation
Notation can be divided naturally into three parts: lower case (generally determinis
tic), upper case (generally random), and Greek.
Lower case
a a (real) parameter
c a constant; coupon rate
dQ
dP
RadonNikodym derivative of Q with respect to P
dt inﬁnitesimal time increment
dW
t
inﬁnitesimal Brownian increment
f a function
f
P
(x) probability density function of the law P
f(t, T) bond forward rates
g a function
g(x, t, T) the function (−log P(t, T)r
t
= x)
i an integer
j an integer
k contract strike/exercise price; an integer; an offset
n an integer
n[t] number of dividend payments made by time t
p, p
j
a probability
q, q
j
a probability
r constant interest rate
r
t
variable interest rate process; instantaneous rate
s initial stock price, alternative time variable
s
j
possible value for the discrete stock process
t time
u foreign currency interest rate; real variable
161
162 APPENDIX B. NOTATION
x a real variable; horizontal axis variable
x
i
(t) timedependent factor of volatility surface
y
i
(T) maturitydependent factor of volatility surface
Upper case
A an event; a constant
A
t
HJM volatility matrix
B
i
, B
t
bond price process
B(t, T) solution of a Riccati equation
C
t
foreign exchange rate; coupon bond price; numeraire
D
i
ﬁnancing gap
D
t
foreign currency cash bond
D(t, T) solution of a Riccati equation
E expectation operator
E
P
expectation under the measure P
E
t
discounted portfolio value process
F forward price
F
s
(t, T) forward price at time s for P(t, T)
F
Q
quanto forward price
F
i
history of discrete stockprice process up to ticktime i
F
t
history of Brownian motion up to time t
I
A
indicator function of the event A
I(t) sequence number of next coupon payment
K option strike price
L(T) LIBOR rate
L(t, T) forward LIBOR rate
M
t
a martingale
N
t
a martingale
N the set of nonnegative integers {0, 1, 2, . . .}
N(µ, σ
2
) a normal random variable with mean µ and variance σ
2
P hypothetical discrete derivative price
P a probability measure
P
T
forward measure
P(t, T) bond prices
Q a probability measure
R
n
the ndimensional real vector space
R(t, T) bond yield surface
163
S
i
, S
t
stock price process
˜
S
t
tradable asset price
S
1
t
. . . S
n
t
stock price processes
T maturity/exercise time of a derivative
T
i
coupon payment times
U
t
foreign currency derivative value process
V derivative value
V
t
derivative value process
V (s, T) BlackScholes option price
W
n
(t) random walk
W
t
Brownian motion
˜
W
t
Brownian motion
W
1
t
, . . . , W
n
t
independent Brownian motions
X random variable; claim value of a derivative
X
i
sequence of random variables
X
t
a stochastic process
Y
t
a stochastic process
Y
i
(t, T) integral of y
i
over [t, T]
Z a (normal) random variable
Z
i
, Z
t
discounted stockprice process
Z(t, T) discounted bond prices
˜
Z
t
discounted tradable asset price
Greek case
α a real parameter
α(t, T) forward rate drift
β(t, T) a function of two variables (Vasicek model)
γ
t
change of measure drift; market price of risk
γ
i
(t, T) BGM volatility surface
δ dividend yield; coupon payment interval
δt a small time increment
δS
i
, δn
i
branch widths
∆S
i
, ∆V
i
change in value across δt of S
i
, V
i
, etc
ζ
t
change of measure process
θ a real variable
θ
t
deterministic drift function
λ a real parameter
164 APPENDIX B. NOTATION
µ constant stock drift
µ
t
variable stock drift process
ν
t
stock drift process
π
i
path probability
Π
i
portfolio
ρ correlation
¯ ρ the orthogonal complement
_
1 −ρ
2
ρ
t
volatility process
σ constant stock volatility
σ
1
, σ
2
stock volatilities
σ
t
variable stock volatility process
σ(t, T) forward rate volatility surface
σ
i
(t, T) multifactor forward rate volatility surface
¯ σ term volatility
Σ
t
volatility matrix
Σ(t, T) bond price volatilities
τ time horizon; maturity date; stopping time
φ
t
,
˜
φ
t
stockholding strategy; representation theorem integrand
Φ normal distribution function: Φ(x) = P
_
N(0, 1) ≤ x
_
ψ
t
bondholding trading strategy
ω a sample path
Appendix C
Glossary of technical terms
Adapted a process which depends only on the current position and
past movements of the driving processes. It is unable to see
into the future
American call option a call option which can be exercised at any time up to the
option expiry date
Arbitrage the making of a guaranteed riskfree proﬁt with a trade or
series of trades in the market
Arbitrage free a market which has no opportunities for riskfree proﬁt
Arbitrage price the only price for a security that allows no arbitrage oppor
tunity
Autoregressive of a process, that it is meanreverting
Average the arithmetic mean of a sample
Bank account process an account which is continuously compounded at the pre
vailing instantaneous rate, and behaves like the cash bond
Binomial process a process on a binomial tree
Binomial representa
tion theorem
a discretetime version of the martingale representation
theorem on the binomial tree
Binomial tree a tree, each of whose nodes branches into two at the next
stage
BlackScholes a stock market model with an analytic option pricing for
mula
Bonds interest bearing securities which can either make regular in
terest payments and/or a lump sum payment at maturity
Bond options an option to buy or sell a bond at a future date
Brownian motion the basic stochastic process formed by taking the limit of
ﬁner and ﬁner random walks. It is a martingale, with zero
drift and unit volatility, and is not Newtonian differentiable
165
166 APPENDIX C. GLOSSARY OF TECHNICAL TERMS
Calculus generally a formal system of calculation, in particular
concerned with analyzing behavior in terms of inﬁnites
imal changes of the variables. Newtonian calculus han
dles smooth functions, but not Brownian motion which re
quires the techniques of stochastic calculus. [Fromcalculus
(Lat.), a pebble used in an abacus]
Call option the option to buy a security at/by a future date for a price
speciﬁed now
CameronMartin
Girsanov theorem
a result which interprets equivalent change of measure as
changing the drift of a Brownian motion
Cap a contract which periodically pays the difference between
current interest rate returns and a rate speciﬁed at the start,
only if this difference is positive. A cap can be used to
protect a borrower against ﬂoating interest rates being too
high
Caplet an individual cap payment at some instant
Cash bond a liquid continuously compounded bond which appreciates
at the instantaneous interest rate
Central Limit theorem a statistical result, which says that the average of a sam
ple of IID random variables is asymptotically normally dis
tributed
Change of measure viewing the same stochastic process under a different set
of likelihoods, changing the probabilities of various events
occurring
Claim a payment which will be made in the future according to a
contract
Commodity a real thing, such as gold, oil or frozen concentrated orange
juice
Complete market a market in which every claim is hedgable
Conditional distribu
tion
the distribution of a random variable conditional on some
information F, such as P(X ≤ xF)
Conditional expecta
tion
taking an expectation given some history as known. For
instance the conational expectation of the number of heads
obtained in three tosses, given that the ﬁrst toss was heads,
is two; whereas the unconditioned expectation is only one
and a half. Written E(·F
t
), for conditioning on the history
of the process up to time t
Contingent claim a claim whose amount is determined by the behavior of
market securities up until the time it is paid
167
Continuous a process or function which only changes by a small amount
when its variable or parameter is altered inﬁnitesimally
Continuoustime a process which depends on a realvalued time parameter,
allowing inﬁnite divisibility of time
Continuously com
pounded
interest is compounded instantly, rather than annually or
monthly, leading to exponential growth
Contract an agreement under law between two principals, or counter
parties
Correlation a measure of the linear dependence of two random vari
ables. If one variable gets larger as the other does, the cor
relation is positive, and negative if one gets larger as the
other gets smaller. The limits of one and minus one corre
spond to exact dependence, whereas independent variables
have zero correlation. Formally correlation is the covari
ance of the random variables divided by the square root of
the product of their individual variances
Coupon a periodic payment made by a bond
Covariance a measure of the relationship of two random variables, the
covariance is zero if the variables are independent (and vice
versa in the case of jointly normal random variables). For
mally the covariance of two variables is the expectation of
their product less the product of their expectations
Cumulative normal in
tegral
see normal distribution function
Currency the monetary unit of a country or group of countries
Default free there being no chance that the bond issuer will be unable to
meet his ﬁnancial undertakings (used theoretically)
Density the probability density function f is the derivative (if it ex
ists) of the distribution function of a continuous random
variable. Intuitively, f(x)dx is the probability that X lies
in the interval [x, x + dx]. The function f is nonnegative,
integrates to one, and can be used to calculate expectations,
and so forth, as
E(X
2
) =
_
∞
−∞
x
2
f(x)dx
Derivative a security whose value is dependent on (derived from) exist
ing underlying market securities. See also contingent claim
168 APPENDIX C. GLOSSARY OF TECHNICAL TERMS
Difference equation the discrete analogue of a differential equation. For exam
ple, to ﬁnd the sequence (x
n
) which obeys
ax
n+2
+ bx
n+1
+ cx
n
= d
Diffusion a stochastic process which is the solution to a SDE
Digital a derivative which pays off a ﬁxed amount if a given future
event happens, and nothing otherwise
Discount scaling a future reward or cost down to reﬂect the impor
tance of now over later
Discount bond a bond which promises to make a lump sum payment at a
future date, but until then is worth less than its face value
Discrete taking distinct, separated values; such as from the sets N or
{0, δt, 2δt, . . .}
Distribution of a random variable, the description of the likelihood of its
every possible value
Distribution function the (cumulative) distribution function F of a random vari
able is deﬁned so that F(x) is the probability that the ran
dom variable is no larger than x. The function F increases
(weakly) from 0 to 1. If F is differentiable, then its deriva
tive is the density
Dividends regular but variable payments made by an equity
Dol´ eans exponential for a local martingale M
t
, this is the solution of the SDE
dX
t
= X
t
dM
t
, which is another local martingale X
t
=
exp
_
M
t
−
1
2
_
t
0
(dM
s
)
2
_
Drift the coefﬁcient of the dt term of a stochastic process
Driftless a process with constant zero drift
Equilibrium distribu
tion
a distribution of a process which is stable under time evolu
tion
Equities stocks which make dividend payments
Equivalent martingale
measure (EMM)
see martingale measure
Equivalent measures two measures P and Q are equivalent if they agree on which
events have zero probability
European call option a call option which can be exercised or not only at the op
tion exercise date. Compare with American call option
Exercise date a set future date at which an option may be exercised or not
Exercise price see strike price
169
Exotics new derivative securities, which will quickly either become
standard products or will sink without trace
Expectation the mean of a random variable, which will be the limiting
value of the average of an inﬁnite number of identical tri
als. For a discrete and a continuous random variable (with
density f) it is respectively
E(X) =
∞
n=0
nP(X = n), E(X) =
_
∞
−∞
xf(x)dx
Exponential Brownian
motion
a process which is the exponential of a drifting Brownian
motion
Exponential martin
gales
the Dol´ eans exponential of a martingale, which itself is a
(local) martingale
Filtration the history, (F
t
)
t≥0
, of a process, where F
t
is the informa
tion about the path of the process up to time t
Fixed of interest rates, that they are constant throughout the term
of the contract
Floating of interest rates, that they can move with the market over
the term of the contract
Floor a contract which periodically pays the difference between
a rate speciﬁed at the start and current interest rate returns,
only if this difference is positive. A ﬂoor can be used to
protect a lender against ﬂoating interest rates being too low.
See also cap
Floorlet which is to ﬂoors as caplets are to caps
Foreign exchange the market which prices one currency in terms of another
Forward an agreement to buy or sell something at a future date for a
set price, called the forward price
Forward rate the forward price of instantaneous borrowing
Fractal a geometrical shape which on a smallscale looks the same
as the largescale, only smaller. A straight line is a fractal
of dimension one, and a Brownian motion path is a fractal
of dimension 1.5
Future a forward traded on an exchange
FX abbreviation for foreign exchange
Gaussian process a process, all of whose marginals are normally distributed,
and all of whose joint distributions are jointly normal
HeathJarrowMorton
(HJM)
a model of the interestrate market
170 APPENDIX C. GLOSSARY OF TECHNICAL TERMS
Hedge to protect a position against the risk of market movements
History the information recording the path of a process
Identically distributed of random variables, have the same probabilistic distribu
tion
IID abbreviation for Independent, Identically Distributed
Independent of variables, none of which have any relation or inﬂuence
on any of the others
Indicator function a function of a set which is one when the argument lies in
the set and zero when it is outside
Induction a method of proof, involving the demonstration that the cur
rent case follows from the previous case, which itself then
implies the next case, and so on
Instantaneous rate the rate of interest paid on a very very short term loan
Instruments tradable securities or contracts
Interest rate the rate at which interest is paid
Interest rate market the market which determines the time value of money
Itˆ o’s formula a stochastic version of the ‘chain rule’ which expresses the
volatility and drift of the function of a stochastic process
in terms of the volatility and drift of the process itself and
the derivatives of the function. If X
t
has volatility σ
t
and
drift µ
t
, then Y
t
= f(X
t
) has volatility f
(X
t
)σ
t
and drift
f
(X
t
)µ
t
+
1
2
f
(t)σ
2
t
Kolmogorov’s strong
law
see strong law
Law of the unconscious
statistician
the result that if a random variable X has density f, then the
expectation of h(X) is
E
_
h(X)
_
=
_
∞
−∞
h(x)f(x)dx
LIBOR the London InterBank Offer Rate. A daily set of interest
rates for various currencies and maturities
Local martingale a stochastic process which is driftless, but not necessarily a
martingale
Logdrift of a stochastic process X
t
, the drift of log X
t
Lognormal distribu
tion
a random variable whose logarithm is normally distributed
Logvolatility of X
t
is the volatility of log X
t
, or equivalently the volatility
of dX
t
/X
t
Long (of position) having a positive holding
171
Marginal the marginal distribution of a process X at time t is the dis
tribution of X
t
considered as a random variable in isolation.
Two processes may be different, yet have exactly the same
marginal distributions
Market a place for the exchanging of price information. Commonly
situated in electronic space
Market maker (in UK) a dealer who is obligated to quote and trade at two
way prices
Market price of risk a standardized reward from risky investments in terms of
extra growth rate
Markov of a process, meaning that its future behavior is independent
of its past, conditional on the present
Martingale a process whose expected future value, conditional on the
past, is its current value. That is, E(M
t
F
s
) equals M
s
for
every s less than t
Martingale measure a measure under which a process is a martingale
Martingale representa
tion theorem
a result which allows one martingale to be written as the
integral of a previsible process with respect to another mar
tingale
Maturity the time at which a bond will repay its principal, or more
generally the time at which any claim pays off
Mean synonym for expectation
Mean reversion the property of a process which ensures that it keeps return
ing to its longterm average
Measure a collection of probabilities on the set of all possible out
comes, describing how likely each one is
Multifactor a market model which i s driven by more than one Brownian
motion
Newtonian calculus classical differential and integral calculus, relating to
smooth or differentiable functions
Newtonian function a function which is smooth enough to have a classical
(Newtonian) derivative
Node a point on a tree where branches start and ﬁnish
Noise a loose term for volatility
Normal distribution a continuous distribution, parameterized by a mean µ and
variance σ
2
, written N(µ, σ
2
) with density
f(x) =
1
√
2πσ
2
exp
_
−
(x −µ)
2
2σ
2
_
172 APPENDIX C. GLOSSARY OF TECHNICAL TERMS
Normal distribution
function
the distribution function of the normal random variable,
written Φ(x) = P(N(0, 1) ≤ X)
Numeraire a basic security relative to which the value of other securi
ties can be judged. Often the cash bond
ODE abbreviation for Ordinary Differential Equation
Option a contract which gives the right but not the obligation to do
something at a future date
OrnsteinUhlenbeck
(OU) process
a mean reverting stochastic process with SDE
dX
t
= σdW
t
+ (θ −αX
t
)dt
Overthecounter an agreement concluded directly between two parties, with
out the mediation of an exchange
Path probability the probability of a tree process taking a particular path
through the tree. The probability will be the product of the
probabilities of the individual branches taken
Payoff a payment
PDE abbreviation for Partial Differential Equation
Poisson process a type of random process with discontinuities
Portfolio a collection of security holdings
Position the amount of a security held, which can either be positive
(a long position) or negative (a short position)
Previsible a stochastic process which is adapted and is either continu
ous or leftcontinuous with rightlimits or is a limit of such
processes
Principal the face value that a bond will pay back at maturity
Probability the chance of an event occurring
Process a sequence of random variables, parameterized by time
Product rule a result giving the stochastic differential of the product of
two stochastic processes
Putcall parity the observation that the worth of a call less the price of a put
struck at the same price is the current worth of a forward
Quantos crosscurrency contracts, derivatives which pay off in an
other currency
RadonNikodym
derivative
of one measure with respect to another is the relative like
lihood of each sample path under one measure compared
with the other
Random variable a function of a sample space
173
Random walk a discrete Markov process made up of the sum of a number
of independent steps. A simple symmetric random walk is
Nvalued and after each time step goes up one with proba
bility
1
2
and down one with probability
1
2
Recombinant tree a tree where branches can come together again
Replicating strategy a selfﬁnancing portfolio trading strategy which hedges a
claim precisely
Risk free no chance of anything going wrong
Riskneutral measure a martingale measure
SDE abbreviation for Stochastic Differential Equation
Security a piece of paper representing a promise
Selfﬁnancing a strategy which never needs to be topped up with extra cash
nor can ever afford withdrawals
Semimartingale a process which can be decomposed into a local martingale
term and a drift term of ﬁnite variation
Share (in UK) a stock or equity
Short (of position) having a negative, or borrowed, holding
Short rate see instantaneous rate
Singlefactor a market model which is driven by only one Brownian mo
tion
Standard deviation the square root of the variance
Stochastic synonym for random
Stochastic calculus a calculus for random processes, such as those involving
Brownian motion terms
Stochastic process a continuous process, which can be decomposed into a
Brownian motion term and a drift term
Stock a security representing partial ownership of a company
Stock market a place for trading stocks
Strike price the price at which an asset may be bought or sold under an
option
Strong law the result that the average of a sample of n IID random vari
ables will converge to the mean of the distribution as n in
creases, given some technical conditions
Swaps an agreement to make a series of ﬁxed payments over time
and receive a corresponding series of payments dependent
on current interest rates, or vice versa
Swaption an option to enter into a swap agreement at a future date
174 APPENDIX C. GLOSSARY OF TECHNICAL TERMS
Taylor expansion for Newtonian functions, the expression of the value of a
function f near x in terms of the value of it and its deriva
tives at x, that is
f(x + h) = f(x) + hf
(x) +
1
2
h
2
f
(x) +
1
6
f
(x) · · ·
Term structure the relationship between the interest rates demanded on
loans, and the length of the loans
Term variance the variance of the logarithm of a security price over a time
period, Var
_
log(S
T
/S
0
)
_
Term volatility the effective (annualized) volatility of an asset over a time
period. Explicitly, its square is the term variance divided
by the length of the term:
¯ σ
2
= Var
_
log(S
T
/S
0
)
_
/T
Time value of money the difference between cash now, and cash later which is
subject to a discount
Tower law the result that E
_
E(XF
t
)F
s
_
= E(XF
s
), for s < t
Tradable of an asset, that it can be traded either directly, or indirectly
by trading a matching portfolio
Trading strategy a continuous choice of portfolio, a choice which may de
pend on market movements
Transaction cost a charge for buying or selling a security
Tree a graph of nodes linked by branches which contains no
closed loops or circuits
Underlying a basic market security, such as stocks, bonds and curren
cies
Vanilla of a product, the standard basic version
Variable coupon periodic payments from a ﬂoating interestrate contract
Variance a measure of the uncertainty of a random variable. For
mally, the expectation of its square less the square of its
expectation, or equivalently the expected square of the dif
ference between the random variable and its mean
Volatility the amount of ‘noise’ or variability of a process, more pre
cisely, the coefﬁcient of the Brownian motion term of a
stochastic process
Weak law the result that the average of n IID random variables is in
creasingly less likely to be signiﬁcantly different from the
distribution mean as n increases
175
Wiener process synonym for Brownian motion
With probability 1 of an event, having probability one of occurring. This is not
quite the same as being guaranteed for sure, as, for example,
a normal random variable can take the value zero, but with
probability one it will not
Yield the average interest rate offered by a bond
Yield curve the graph of yield plotted against bond maturity
Zero coupon a bond which does not make any payments until maturity
Contents
Preface The parable of the bookmaker 1 Introduction 1.1 Expectation pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Arbitrage pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Expectation vs arbitrage . . . . . . . . . . . . . . . . . . . . . . . . . 2 Discrete processes 2.1 The binomial branch model . . . 2.2 The binomial tree model . . . . 2.3 Binomial representation theorem 2.4 Overture to continuous models . i iii 1 1 4 5 7 7 12 22 32 34 34 39 44 48 59 61 64 71 77 77 83 87 90 95
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
3 Continuous processes 3.1 Continuous processes . . . . . . . . . . . . 3.2 Stochastic calculus . . . . . . . . . . . . . 3.3 Itˆ calculus . . . . . . . . . . . . . . . . . . o 3.4 Change of measure — the CMG theorem 3.5 Martingale representation theorem . . . . . 3.6 Construction strategies . . . . . . . . . . . 3.7 BlackScholes model . . . . . . . . . . . . 3.8 BlackScholes in action . . . . . . . . . . . 4 Pricing market securities 4.1 Foreign exchange . . . 4.2 Equities and dividends 4.3 Bonds . . . . . . . . . . 4.4 Market price of risk . . 4.5 Quantos . . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
5 Interest rates 100 5.1 The interest rate market . . . . . . . . . . . . . . . . . . . . . . . . . . 100 5.2 A simple model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
ii
CONTENTS
i
5.3 5.4 5.5 5.6 5.7
Singlefactor HJM . . Shortrate models . . Multifactor HJM . . Interest rate products Multifactor models .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
111 117 123 127 134 139 139 141 143 147 150 153 157 161 165
6 Bigger models 6.1 General stock model . . . . . . . . . . 6.2 Lognormal models . . . . . . . . . . 6.3 Multiple stock models . . . . . . . . . 6.4 Numeraires . . . . . . . . . . . . . . . 6.5 Foreign currency interestrate models 6.6 Arbitragefree complete models . . . A Further reading B Notation C Glossary of technical terms
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
Chapter three repeats all the work of its predecessor in the continuous time setting. Such ‘seat of the pants’ practices are more suited to the pioneering days of an industry. Brownian motion is brought out. the importance of hedging both as a justiﬁcation for the price and as an important end in itself is often underplayed. This is usually enough for experienced practitioners to price existing contracts. as well as the Itˆ calculus needed to manipulate it. The key probabilistic concepts of conditional expectation. and much market practice is based on a soft understanding of what is actually going on. Scholars need to be aware of such ﬁnancial issues. martingales. When working in isolation from the market. In particular. but often insufﬁcient for innovative new products.
Guide to the chapters
Chapter one is a brief warning. the temptation is to ﬁnd analytic answers for their own sake with no reference to the concerns of practitioners. such as dividend paying equities. managers and regulators can be left to stumble around in literature which is ill suited to their need for a clear explanation of the basic principles. That idea has to be shaken off and arbitrage pricing take its place. o culminating in a derivation of the BlackScholes formula. and representation are all introduced in this simple framework. effort is too often expended on ﬁnding precise answers to the wrong questions.Preface
Notoriously. accompanied by illustrative examples. A general pattern of the distinction between tradi
. especially to beginners. Novices. The mathematics of ﬁnance is not easy. Chapter two develops the idea of hedging and pricing by arbitrage in the discretetime setting of binary trees. But both are needed. currencies and coupon paying bonds. as Dr Johnson might have put it. and those comprehensible are not necessarily precise. Sadly. rather than the mature $15 trillion market which the derivatives business has become. the ones which are precise are not necessarily comprehensible. that the expected worth of something is not a good guide to its price. and adapts the BlackScholes approach to each in turn. if only because some of the very best work has arisen in answering the questions of industry rather than academe. Chapter four runs through a variety of actual ﬁnancial instruments. works of mathematical ﬁnance can be precise. change of measure. On the academic side. and they can be comprehensible.
The aim is to tell one basic story of the market. which lies within the general continuous framework set up in chapter three. as well as a warning not to take that name too seriously. the many readers in London. as well as his much more invaluable positive assistance. a full glossary of technical terms and an index are in the appendices. The running link between the existence of equivalent martingale measures and the ability to price and hedge is ﬁnally formalized. Chapter six concludes with some technical results about larger and more general models. The occasional questions are to allow practice of the requisite skills. or by random access to the selfcontained sections. including swaps. but the richness of this market makes it more than just a special case of BlackScholes. caps/ﬂoors and swaptions. A short bibliography. This is a substantial chapter reﬂecting the depth of ﬁnancial and technical knowledge that has to be introduced in an understandable way. which all approaches can slot into. In spirit. a market of bonds is much like a market of stocks. A section on quanto products provides a showcase of examples. complete answers to the (small) number of exercises. and are never essential to the development of the material.
Acknowledgements
We would like to thank David Tranah at CUP for politely never mentioning the number of deadlines we missed. except for some (classical) differential calculus and experience with symbolic notation. Chapter ﬁve is about the interest rate market. One section details a few of the many possible interest rate contracts. Some basic probability deﬁnitions are contained in the glossary. A reader is not expected to have any particular prior body of knowledge. stochastic numeraires. Special thanks to Lorne Whiteway for his help and encouragement.ii
PREFACE
able and nontradable quantities leads to the deﬁnition the market price of risk. whereas more advanced readers will ﬁnd technical asides in the text from time to time. June 1996 Martin Baxter Andrew Rennie
. and foreign exchange interestrate models. New York and various universities who have been subjected to writing far worse than anything remaining in the ﬁnished edition. including multiple stock nfactor models.
How to read this book
The book can be read either sequentially as an unfolding story. Market models are discussed with a joint shortrate/HJM approach.
The parable of the bookmaker
A
bookmaker is taking bets on a twohorse race. Choosing to be scientiﬁc, he studies the form of both horses over various distances and goings as well as considering such factors as training, diet and choice of jockey. Eventually he correctly calculates that one horse has a 25% chance of winning, and the other a 75% chance. Accordingly the odds are set at 31 against and 31 on respectively. But there is a degree of popular sentiment reﬂected in the bets made, adding up to $5 000 for the ﬁrst and $10 000 for the second. Were the second horse to win, the bookmaker would make a net proﬁt of $1667, but if the ﬁrst wins he suffers a loss of $5000. The expected value of his proﬁt is 25% × (−$5000) + 75% × ($1667) = $0, or exactly even. In the long term, over a number of similar but independent races, the law of averages would allow the bookmaker to break even. Until the long term comes, there is a chance of making a large loss. Suppose however that he had set odds according to the money wagered — that is, not 31 but 21 against and 21 on respectively. Whichever horse wins, the bookmaker exactly breaks even. The outcome is irrelevant. In practice the bookmaker sells more than 100% of the race and the odds are shortened to allow for proﬁt (see table). However, the same pattern emerges. Using the actual probabilities can lead to longterm gain but there is always the chance of a substantial shortterm loss. For the bookmaker to earn a steady riskless income, he is best advised to assume the horses’ probabilities are something different. That done, he is in the surprising position of being disinterested in the outcome of the race, his income being assured. A note on odds When a price is quoted in the form nm against, such as 31 against, it means that a successful bet of $m will be rewarded with $n plus stake returned. The implied probability of victory (were the price fair) is m/(m + n). Usually the probability is less than half a chance so the ﬁrst number is larger than the second. Otherwise, what one might write as 13 is often called odds of 31 on.
iii
iv
THE PARABLE OF THE BOOKMAKER
Actual probability Bets placed 1. Quoted odds Implied probability Proﬁt if horse wins 2. Quoted odds Implied probability Proﬁt if horse wins
25% 75% 135 against 28% $3000 95 against 36% $1000
$5000 $10 000 154 on 79% $2333 52 on 71% $1000 Total = 107% Expected proﬁt = $1 000 Total = 107% Expected proﬁt = $1 000
Allowing the bookmaker to make a proﬁt, the odds change slightly. In the ﬁrst case, the odds relate to the actual probabilities of a horse winning the race. In the second, the odds are derived from the amounts of money wagered.
Chapter 1
Introduction
F
inancial market instruments can be divided into two distinct species. There are the ‘underlying’ stocks: shares, bonds, commodities, foreign currencies; and their ‘derivatives’, claims that promise some payment or delivery in the future contingent on an underlying stock’s behavior. Derivatives can reduce risk — by enabling a player to ﬁx a price for a future transaction now, for example — or they can magnify it. A costless contract agreeing to pay off the difference between a stock and some agreed future price lets both sides ride the risk inherent in owning stock without needing the capital to buy it outright. In form, one species depends on the other — without the underlying (stock) there could be no future claims — but the connection between the two is sufﬁciently complex and uncertain for both to trade ﬁercely in the same market. The apparently random nature of stocks ﬁlters through to the claims — they appear random too. Yet mathematicians have known for a while that to be random is not necessarily to be without some internal structure — put crudely, things are often random in nonrandom ways. The study of probability and expectation shows one way of coping with randomness and this book will build on probabilistic foundations to ﬁnd the strongest possible links between claims and their random underlying stocks. The current state of truth is, however, unfortunately complex and there are many false trails through this zoo of the new. Of these, one is particularly tempting.
1.1 Expectation pricing
Consider playing the following game — someone tosses a coin and pays you one dollar for heads and nothing for tails. What price should you pay for this prize? If the coin is fair, then heads and tails are equally likely — about half the time you should win the dollar and the rest of the time you should receive nothing. Over enough plays, then, you expect to make about ﬁfty cents a go. So paying more than ﬁfty cents seems extravagant and less than ﬁfty cents looks extravagant for the person offering the game. Fifty cents, then, seems about right. Fifty cents is also the expected proﬁt from the game under a more formal, mathematical deﬁnition of expectation. A probabilistic analysis of the game would observe
1
but instead of allowing you a number of plays. and the number of ‘buyers’ or ‘sellers’ happy with that price might have nothing to do with the mathematical expectation of the game’s outcome. In the short term of course. that is Sn = (X1 + X2 + . Suppose instead that the coin game took place at the end of a year. as n gets larger the value of Sn tends towards the mean µ of the distribution. all sampled from the same distribution. The strong law lets you take advantage of them over repeated plays: 40 cents a dollar would then be ﬁnancial suicide. INTRODUCTION
that although the outcome of each coin toss is essentially random. but over time. X3 . but it does nothing if you are allowed just one play. in this discrete case. but payment to play had to be made at the beginning — in effect we had to ﬁnd the
. and so on. And along with a probability ascription comes the idea 2 of expectation. then in the long run it is certain that you will end up in proﬁt. X2 . Fifty cents is a fair price in this sense. If the arithmetical average of outcomes tends towards the mathematical expectation with certainty. and we let Sn be the arithmetical average of the sequence up to the nth term. with probability one. a probability of the coin landing heads or tails of 1 . gave you just one for an arbitrarily large payoff. + Xn )/n. . And if it is negative. Our analysis of the coin game was simpliﬁed by the payment for and the payoff from the game occurring at the same time. then the average proﬁt/loss per game tends towards the mathematical expectation less the price paid to play the game. then you will approach an overall loss with certainty. But as a guide to a starting price for the game. the strong law coupled with expectation seems to have something going for it. So the ‘market’ in this game could trade away from an expectation justiﬁed price. a ballpark amount to charge. Then. which has mean (expectation) µ. . But is it an enforceable price? Suppose someone offered you a play of the game for 40 cents in the dollar. this is not inconsistent with a deeper nonrandom structure to the game. nothing can be guaranteed.
Time value of money
We have ignored one important detail — the time value of money.50. selling off all your belongings and taking out loans to the limit of your credit rating would not be a rational way to take advantage of this source of free money. We could posit that there was a ﬁxed measure of likelihood attached to the coin tossing. expectation will out. the total of each outcome’s value weighted by its attached probability. Mortgaging your house. Any price might actually be charged for the game in the short term. The expected payoff in the game is 1 × $1 + 1 × $0 = $0. 2 2 This formal expectation can then be linked to a ‘price’ for the game via something like the following: Kolmogorov’s strong law of large numbers Suppose we have a sequence of independent random numbers X1 .2
CHAPTER 1. If this difference is positive.
that is when the stock transfer actually takes place. As with the time value of money above. The ‘pricing question’ for the forward stock ‘game’ is: what amount should be written into the contract now to pay for the stock one year in the future? We can dress this up in formal notation — the stock price at time T is given by ST . We could assume the existence of a market for these future promises.1. that we have some claim on this stock. but something less. The oldest and possibly most natural claim on a stock is the forward: two parties enter into a contract whereby one agrees to give the other the stock at some agreed point in the future in exchange for an amount agreed now. and the forward payment written into the contract is K . thus the value of the contract at its expiry. the value now of a dollar promised at time T is given by exp(−rT ) for some constant r > 0. Paying 50 cents at time T is the same as paying 50 exp(−rT ) cents now.
Suppose. EXPECTATION PRICING
3
value of the coin game’s contingent payoff not as of the future date of play. r doesn’t have to be constant. If we are in January. not coins
What about real stock prices in a real ﬁnancial market? One widely accepted model holds that stock prices are lognormally distributed. some contract that agrees to pay certain amounts of money in certain situations — just as the coin game did. is
. promise) now. now. We can derive a stronglaw price for the game played at time T . for cost 50 exp(−rT ) cents. Interest rates are the formal acknowledgement of this. and bonds are the market derived from this. That is
log ST = log S0 + X
or
ST = S0 exp(X). But here we assume it is. the prices quoted for these bonds being structured. we should formalize this belief. such that the change in the logarithm of the stock price over some time period T is given by X . then one dollar in December is not worth one dollar now.
Stocks.1. Why? Because the payment of 50 cents at time T can be guaranteed by buying half a unit of the appropriate bond (that is. Thus the stronglaw price must be not 50 cents but 50 exp(−rT ) cents. The stock is being sold forward. Speciﬁcally: Time value of money We assume that for any time T less than some time horizon r. The rate r is then the continuously compounded interest rate for this period. derivable from some interest rate. Stock model We assume the existence of a random variable X . which is normally distributed with mean µ and standard deviation σ . but as of now. The interest rate market doesn’t have to be this simple. And indeed in real markets it isn’t.
The fair price of the contract is S0 exp(rT ).4
CHAPTER 1. Why does the strong law fail so badly with forwards? As mentioned above in the context of the coin game.
then longterm use of that pricing should lead to one side’s proﬁt.
Integration and the law of the unconscious statistician then tells us that the expected stock price at time T is S0 exp µ + 1 σ 2 . it is also completely useless. It doesn’t depend on the expected value of the
. But the technique will clearly work for more than just forwards. so plausible — but seductive though the strong law is. should be zero. which happens when K = E(ST ). then for any integrable real function h. another completely different mechanism does enforce a price. The price we have just determined for the forward could only be the market price by an unfortunate coincidence. This is the stronglaw justiﬁed price for 2 the forward contract. the expectation of h(X) is
∞
E(h(X)) =
−∞
h(x)f (x)dx.
1. The time value of money tells us that the value of this claim as of now is exp(−rT )(ST − K).2 Arbitrage pricing
So far. h(X). Discounting this expectation then gives a theoretical value which the strong law tempts us into linking with economic reality. Many claims are capable of translation into functional form. Thus one apparently reasonable answer to the pricing question says K should be set so that E(exp(−rT )(ST − K)) = 0. And in this case. it only suggests. and the law of the unconscious statistician should be able to deliver an expected value for them. the strong law cannot enforce a price. The strong law suggests that the expected value of this random amount. E(exp(−rT )(ST − K)). With markets where the stock can be bought and sold freely and arbitrary positive and negative amounts of stock can be maintained without cost. What is E(ST )? We have assumed that log(ST ) − log(S0 ) is normally distributed with mean µ and variance σ 2 — thus we want to ﬁnd E(S0 exp(X)). INTRODUCTION
ST − K . it can only be a suggestion as to the market’s trading level. trying to trade forward using the strong law would lead to disaster — in most cases there would be unlimited interest in selling forward to you at that price. If it is positive or negative. just as with the coin game. we can use a result such as: The law of the unconscious statistician Given a realvalued random variable X with probability density function f (x).
Since X is normally distributed. For that. where X is normally distributed with mean µ and standard deviation σ . the probability density function for X is
f (x) = √ 1 2πσ 2 exp −(x − µ)2 2σ 2 .
If the stock price ends up above the strike.3. Thus there is an enforced price. They could borrow S0 now. And unlike the coin game. So the forward price is bounded below by S0 exp(rT ). But if the stock price ends up below the strike. the forward is a special case. Almost everything appeared safe to price via
. thus writing more than S0 exp(rT ) into the contract would guarantee them a loss. it doesn’t even depend on the stock price having some particular distribution. buy the stock with it. overrides the strong law. But of course. The forward price is bounded above 1 by S0 exp(rT ) as well. When the contract expires. the buyer of the contract can run the scheme in reverse. But in a certain sense. The price of S0 exp(rT ) is an arbitrage price — it is justiﬁed because any other price could lead to unlimited riskless proﬁts for one party. Any attempt to strike a different price and offer it into a market would inevitably lead to someone taking advantage of the free money available via the construction procedure. the buyer will abandon the option and any stock owned by the seller would have incurred a pointless loss. they have to pay back the loan — which if the continuously compounded rate is r means paying back S0 exp(rT ). obliged to deliver the stock at time T in exchange for some agreed amount. This type of market opportunism is old enough to be ennobled with a name — arbitrage.1. however surprising. mortgaging the house would now be a rational action. any other price is too dangerous to quote.
Construction strategy
Consider the seller of the contract. The standard call option which offers the buyer the right but not the obligation to receive the stock for some strike price agreed in advance certainly couldn’t be constructed this way. put the stock in a drawer and just wait. not of S0 exp µ + 2 σ 2 but S0 exp(rT ). EXPECTATION VS ARBITRAGE
5
stock.) But the existence of an arbitrage price. then the buyer would exercise the option and ask to receive the stock — having it salted away in a drawer would then be useful to the seller. If they wrote less than S0 exp(rT ) into the contract as the amount for forward payment.3 Expectation vs arbitrage
The strong law and expectation give the wrong price for forwards. and until 1973. Thus maybe a stronglaw price would be appropriate for a call option. (But then of course. then a buyer of a forward contract expects to make money. then they would lose money with certainty. To put it simply. The construction strategy — buying the stock and holding it — certainly wouldn’t work for more complex claims. so would buyers of the stock itself. but they have the stock ready to deliver. many people would have agreed. The strong 1 law wasn’t wrong — if S0 exp µ + 2 σ 2 is greater than S0 exp(rT ). if the stock is expected to grow faster than the riskless interest rate r. if there is an arbitrage price.
1. Either counterparty to the contract can in fact construct the claim at the start of the contract period and then just wait patiently for expiry to exchange as appropriate.
Since 1973. Nowhere in this book will we use the strong law again. though. and the infamous BlackScholes paper. however. and only forwards and close relations seemed to have an arbitrage price.
. but it will be as a tool for riskfree construction. just how wrong this is has slowly come out. All derivatives can be built from the underlying — arbitrage lurks everywhere.6
CHAPTER 1. Just to muddy the waters. INTRODUCTION
expectation and the strong law. expectation will be used repeatedly.
2.
The stock
Just one timetick — we start at time t = 0 and end a short tick later at time t = δt.1). With just two things allowed to happen. We have a long way to go.1: The binomial branch
Our randomness will have some structure — we will assign probabilities to the up
7
. it seems wise to start very small. then. Bit by bit we will put together a mathematical framework strong enough to be a realistic model of the real ﬁnancial markets and yet still structured enough to support construction techniques.1 The binomial branch model
Something random for the stock and something to represent the timevalue of money. We need something to represent the stock. some random component.
Figure 2. Consider. the simplest possible model with a stock and a bond.Chapter 2
Discrete processes
T
he goal of this book is to explore the limits of arbitrage. pictorially we have a branch (ﬁgure 2. At the very least we need these two things — any model without them cannot begin to claim any relation to the real ﬁnancial market. and it had better have some unpredictability. though. So we suppose that only two things can happen to the stock in this time: an ‘up’ move or a ‘down’ move.
Nothing happens to us in the intervening period by dint of holding on to the stock — there is no charge for holding positive or negative amounts — but at the end of the period it will have a new value. and sell off cash bonds to fund the purchase. The investor’s requirement for compensation based on the future value of the stock could be codiﬁed by a function f mapping the two future possibilities. struck at k . The stock will have some value at the start (node 1 as labeled on the picture). up. f (3) = s3 − k . say B0 .8
CHAPTER 2. we introduce a cash bond B which we can buy or sell at time zero for some price. we would buy the stock (cost: s1 ).
Riskfree construction
The question can now be posed — exactly what class of functions f can be explicitly constructed via a suitable strategy? Clearly the forward can be — as in chapter one. To represent this. to two rewards or penalties f (2) and f (3). to node 3. The probabilities of each outcome for the derivative f are known. and simple though it is it still has uncertainties for investors. But what about more complex f ? Can we still ﬁnd a construction strategy? Our ﬁrst guess would be no. but we don’t know its actual value in advance. node 2 or node 3. This value represents a price at which we can buy and sell the stock in unlimited amounts. and which will be worth a deﬁnite B0 exp(rδt) a tick later. We can then hold on to the stock across the time period until time t = δt. then it will have value s2 . DISCRETE PROCESSES
and down move: probability p to move up to node 3. to node 2. we would be able to hand over the stock and demand s1 exp(rδt) in exchange. The stock takes one of two random values at the end of the tickperiod and the value of the derivative would in general be different as well. and thus 1 − p to move down to node 2. for example.
. and borrow — and in arbitrary size. At the end of the period. These two instruments are our ﬁnancial world. The price k of the forward thus has to be s1 exp(rδt) exactly as we would have hoped — priced via arbitrage. A forward contract. Only one of the possible stock values might suit a particular player. Thus there could be a market for instruments dependent on the value the stock takes at the end of the tickperiod. their plans surviving or failing by the random outcome. We should be able to lend at that rate. There will be some continuously compounded interest rate r that will hold for the period t = 0 to t = δt — one dollar at time zero will grow to $ exp(rδt). thus we also know the expected value of f at the end of the period as well: (1 − p)f (2) + pf (3).
The bond
We also need something to represent the timevalue of money — a cash bond. If it moves down. value s3 . could be codiﬁed as f (2) = s2 − k . call it s1 .
downvalue s2 and upvalues s3 . And we already know that this best isn’t good enough for forwards. and we chose the expected value of the derivative as our target point. This strategy of construction would at the very least be expected to break even. But of course this is just the strong law of chapter one all over again — just thinly disguised as construction. And exactly as before we are missing an element of coercion. we have after all two instruments which we can build into a portfolio to hold for the tickperiod. The discounted expectation of the claim doesn’t work as a pricing tool. Why would we choose this value as the target to aim for? Because it is the expected value of the derivative at the end of the period — formally: Expectation for a branch Let S be a binomial branch process with base value s1 at time zero. but enforced by the interest rate r implied by the cash bond B : namely s1 exp(rδt). Namely the stock itself. Then the expectation of S at ticktime 1 under the probability of an upmove p is:
Ep (S1 ) = (1 − p)s2 + ps3
Our claim f on S is just as much a random variable as S1 is — we can meaningfully talk of its expectation. but the value of the derivative
.2. The price we would predict for the derivative would be the discounted expectation of its value at the end. its forward price is not suggested by the possible stock values s2 and s3 . though. thus buying discount bonds to the value of exp(−rδt)[(1 − p)f (2) + pf (3)] at the start of the period will provide a value equal to (1 − p)f (2) + pf (3) at the end. via the cash bonds. we have simply aimed between them in a probabilistic sense and hoped for the best. We haven’t explicitly constructed the two possible values the derivative can take: f (2) and f (3).
Stocks and bonds together
But can we do any better? Another strategy might occur to us. And the value of the starting portfolio of cash bonds might be claimed to be a good predictor of the value of the derivative at the start of the period. THE BINOMIAL BRANCH MODEL
9
Bondonly strategy
All is not lost.1. For a stock that obeys a binomial branch process. The cash bond will grow by a factor of exp(rδt) across the period. Consider a portfolio of just the cash bond. But we have another instrument tied more strongly to the behavior of both the stock and the derivative than just the cash bond. Suppose we attempted to guarantee not an amount known in advance which we hope will stand as a reasonable predictor for the value of the derivative. And thus we can meaningfully aim for the expectation of the claim. We tried using the guaranteed growth of the cash bond as a device for producing a particular desired value.
whatever the stock
. two possible claim values and two free variables φ and ψ . ψ). ψ) portfolio and held it. We have synthesized the derivative. We have two values f (3) and f (2) which we want to duplicate under the appropriate move of the stock. Anyone could buy the derivative from them in arbitrary quantity. ψ) portfolio. Any derivative f can be constructed from an appropriate portfolio of bond and stock. ψ) portfolio to exactly match it. it would cost φs1 + ψB0 . φs2 + ψB0 exp(rδt) = f (2). And constructed in advance. though. This pair of equations should intrigue us — we have two equations.
and φs2 + ψB0 exp(rδt) after an ‘down’ move. then the portfolio becomes worth f (3). One tick later. and sell the (φ.10
CHAPTER 2. and if the stock moves down. and we have two variables φ and ψ which we can adjust. this is enforceable in an ideal market as a rational price. the equations guarantee that we achieve our goal — if the stock moves up. Consider a general portfolio (φ. whatever it might be. s3 − s2 (f (3) − f (2))s3 s3 − s2 . At the end of the tickperiod the value of the derivative would exactly cancel the value of the portfolio. and of course it does — unlike the expectation derived value.
Except if perversely s2 and s3 are identical — in which case S is a bond not a stock — we have the solutions:
φ= f (3) − f (2) . namely φ of the stock S (worth φs1 ) and ψ of the cash bond B (worth ψB0 ). Thus the strategy can reduce to solving the following two simultaneous equations for (φ. Denote by V the value of buying the (φ. This must have some effect on the value of the claim. namely φs1 + ψB0 . which is:
V = s1 f (3) − f (2) s3 − s2 + exp(−rδt) f (3) − (f (3) − f (2))s3 s3 − s2
Now consider some other market maker offering to buy or sell the derivative for a price P less than V . If we were to buy this portfolio at time zero.
−1 ψ = B0 exp(−rδt) f (3) −
What can we do with this algebraic result? If we bought this (φ. the portfolio becomes worth f (2). DISCRETE PROCESSES
itself. ψ):
φs3 + ψB0 exp(rδt) = f (3).
The price is right
Our simple model allows a surprisingly prescient strategy. it would be worth one of two possible values:
φs3 + ψB0 exp(rδt) after an ‘up’ move.
But with an eye to rearranging the equations.2. Similarly if a market maker quoted the derivative at a price P greater than V . The picture describes the situation:
Figure 2. After a downjump. this portfolio becomes worth 2 × $2 − 3 × $1 = $1. lets arbitrage creep everywhere — the strong law can be banished completely. The cost of this portfolio at time zero is 2 × $1 − 1 × $1 = $0. and X the payoff of the bet. anyone could make arbitrary riskfree proﬁts. What is the worth of a bet which pays $1 if the stock goes up? Solution. Let B denote the bond price.50. 3 it is worth 2 × $0.33. ψ) portfolio to lock in a riskfree proﬁt of P − V per unit trade.2: Pricing a bet
Buy a portfolio consisting of 2/3 of a unit of stock and a borrowing of 1/3 of a unit of bond. though allowing randomness. It must be that the portfolio’s initial value of $0. Only by quoting a twoway price of V can the market maker avoid handing out riskfree proﬁts to other players — hence V is the only rational price for the derivative at time zero. S the stock price. anyone could sell them it and buy the (φ. Our model. At the end of the next time interval. The portfolio exactly simulates the bet’s payoff.5 − 1 × $1 = $0. the stock is worth either $2 or $0. But after 3 3 1 an upjump. the start of the tickperiod.
Example — the whole story in one step
We have an interestfree bond and a stock. expectation pricing involving the probabilities p and 1 − p leads to riskfree proﬁts being available. both initially priced at $1. The stronglaw approach may be useless in this model — leaving aside coincidence. So P would not have been a rational price for the market maker to quote and the market would quickly have mobilized to take advantage of the ‘free’ money on offer in arbitrary quantity. Again the market would take advantage of the opportunity.
Expectation regained
A surprise still lurks.33 is also the bet’s initial value. THE BINOMIAL BRANCH MODEL
11
price was — thus this set of trades carries no risk.
. 3 3 so has to be worth exactly the same as the bet. But the trades were carried out at a proﬁt of V −P per unit of derivative/portfolio bought — by buying arbitrary amounts.1.
12
CHAPTER 2. DISCRETE PROCESSES
we can deﬁne a simplifying variable:
q= s1 exp(rδt) − s2 . then we could transfer its results into a larger. better model. It sounds pedantic to say it. s3 − s2
What can we say about q ? Without loss of generality. but V is an expectation. Similarly were q to be greater than or equal to 1. where f (2) = s2 − k and f (3) = s3 − k . that the correct strike price is indeed s1 exp(rδt). can be thought of as the payoff f . the interval of points strictly between 0 and 1 — the same constraint we might demand for a probability. But s1 exp(rδt) is the value that would be obtained by buying s1 worth of the cash bond B at the start of the tickperiod. Thus the stock could be bought in arbitrary quantity ﬁnanced by selling the appropriate amount of cash bond and a guaranteed riskfree proﬁt made. The price V is not the expected future value of the derivative (discounted suitably by the growth of the cash bond) — that would involve p in the above formula. not an expected value. but it represents a bare minimum as a model. So for any market in which we have a stock which obeys a binomial branch process S . struck at k . this is the expectation of the claim under q . And it is easy enough to check that this expectation gives the correct strike for a forward contract: s1 exp(rδt). Thus the structure of a rational market will force q into (0. Now verify. we have q > 0. we can assume that s3 is bigger than s2 . Our single time step was simple to analyze. Were q to be less than or equal to 0. If we view the expectation operator as implying some information about the future — a stronglaw average over many trials. using the formula for V . 1).
xxxx. then s2 < s3 ≤ s1 exp(rδt) — and this time selling stock and buying cash bonds provides unlimited riskfree gains. It had a random stock and a cash bond.
2. Markets are not quite that straightforward. This reappearance of the expectation operator is unsettling. for example — then V is not what we would unconsciously call the expected value.2 The binomial tree model
From branch to tree. Yet V is the discounted expectation with respect to some number q in (0. Now the surprise: when we rewrite the formula for the value V of the (φ.
Outrageous though it might seem.1 Show that a forward contract. but it only allowed the stock two possible values at the end of a single time period. It is not unreasonable then to eliminate this possibility by ﬁat — specifying the structure of our market to avoid it. ψ) portfolio (try it) we get:
V = exp(−rδt) (1 − q)f (2) + qf (3) . 1). But if we could build the branch model up into something more sophisticated. then s1 exp(rδt) ≤ s2 < s3 . TNxxx
Exercise 2. This is
.
then the actual value one tick later is not known but the range of possibilities is −S1 has only two possible values: s2 and s3 . Now. from t = 0 to t = δt. but dependent on the value at ticktime 1. Our ministock from the binomial branch model allowed the stock to change to just two values at the end of the time period. or bidoffer spreads. We also have a richer allowed structure of probability. will be just as before (a tree of branches starts with just one simple branch). Unlimited amounts of either can be bought and sold without transaction costs. and we shall keep that structure. S2 can be either s4 or s5 . and see what survives. A claim can now call on not just two possibilities. One tick δt later still. As the picture suggests.3
This tree arrangement gives us considerably more ﬂexibility. from s3 . If the value of S at time zero is S0 = s1 . Our ﬁnancial world will again be just two instruments — a discount bond B and a stock S . at ticktime i. but any number.
Figure 2. we will allow many. there are still only two admissible possibilities: from node j the process either goes down to node 2j or up to node 2j +1. stringing the individual δts together. But now.2. though of course given the value at ticktime (i−1). the stock again has two possibilities. The very ﬁrst time period. then we merely have to set δt small enough that we get ten or so layers of the tree in before the claim time t. we must extend the branch idea in a natural fashion. instead of a single timeperiod. S2 can be either s6 or s7 . THE BINOMIAL TREE MODEL
13
the intention of this section — we shall build a tree out of branches.2. From s2 .
The stock
Changes in the value of the stock S must be random — the market demands that — but the randomness can have structure. the stock can have one of 2i possible values. default risks. we will string these choices together into a tree. If we think that a thousand random possible values for a stock is a suitable level of complexity. Each up/down choice will
. hence there are four possibilities. But now.
Trees are complex
At this stage. We have admitted the possibility of randomness in the cash bond’s behavior (though in fact we will not yet be particularly interested in its exact form). given its previous value of sj . From the standpoint of notation. each known at the start of the approprin−1 ate tick period. DISCRETE PROCESSES
have an attached probability of it being made. There is no reason to impose such a strict condition — we don’t have to have a constant interest rate known for the entire tree in advance but instead we could have a sequence of interest rates. the binomial structure of the tree may seem rather arbitrary. In the simple branch model. we need a grownup cash bond. before we abandon the tree as simplistic.
The cash bond
To go with our grownup stock. the cash bond behaved entirely predictably. an understanding of the limitations (or lack of them) of riskfree construction when the underlying stocks take continuous values in continuous time. A tree is better than a single branch. and achieving value s2j . But compared to the stock it is a very different sort of randomness. but it still won’t allow continuous ﬂuid changes in stock and bond values.1 to an entire binomial tree is
. as we shall see. Again this is shown in the picture. and in this case the price of the cash bond at time nδt is B0 exp(rnδt). is then 1 − pj . we had better check that it hasn’t become too complex for us to make any analytic progress at all. but the value of a cash holding at the next tick point is always known. we can represent this pair of probabilities (which must sum to 1) by just one of them (the up probability) pj . . In fact. the probability of the stock achieving value s2j+1 .. this model will in fact be more than capable of matching the models we have in mind.14
CHAPTER 2. R0 . Our ﬁnal goal. The cash bond B has the same structure as the time value of money. we will now keep a constant interest rate r applying everywhere in the tree. it more than suits our purpose. Perhaps more pertinently. will draw directly and naturally on this starting point. The interest which must be paid or earned on cash can change over time. It is worth contrasting the cash bond and the stock. or indeed unnecessarily simplistic.
Backwards induction
In fact most of the hard work has already been done when we examined the branch model. The value of the cash bond at time nδt thus be B0 exp i=0 Ri δt . . . Extending the results and intuitions of section 2. there was a known interest rate r which applied across the period making the cash bond price increase by a factor of exp(rδt). The probability of the stock moving down. because it depends only on the interest rate already known at the start of the period. R1 . But for simplicity’s sake. And as δt tends to zero.
three branches stuck together into a tree. a general claim for our stock S .
The twostep
We know that the expectation operator can be made to work for a single branch — here. we must wade through the algebra for two timesteps. THE BINOMIAL TREE MODEL
15
surprisingly straightforward. then.2. Consider. No other history reaches that node. When we examined a single branching of our tree. are both structurally identical to the simple
. So in our twostep tree (ﬁgure 2.4: Double fork at time 0
Suppose that the interest rate over any branch is constant at rate r.
Figure 2. There must be some ﬁnal ticktime at which the claim is fully determined. the two forks from node 3 to nodes 6 and 7. A general claim can be thought of as some function on the nodes at this claim timehorizon. The tree structure of the stock was not entirely arbitrary — it embodies a onetoone relationship between a node and the history of the stock’s path up to and including that node. then. The key idea is that of backwards induction — extending the construction portfolio back one tick at a time from the claim to the required starting place. A condition not unreasonable in the real ﬁnancial world. and trivially no other node is reached by that history. If two timesteps work. we had the function f dependent only on the node chosen at the end of a single tick period — here we can extend the idea of a claim to cover not only the value of S at the time the claim is exercised but also the history of S up until that point. We shall also insist on the ﬁniteness of our tree.2. Then there exists some set of suitable qj s such that the value of the derivative at node j at ticktime i. is
f (j) = e−rδt qj f (2j + 1) + (1 − qj )f (2j) . and from node 2 to nodes 4 and 5.4).
That is the discounted expectation under qj of the time — (i+1) claim values f (2j +1) and f (2j). f (j). then so will many. This is precisely that condition that allows us actually to associate a claim value with a particular endnode on our tree.
f (2) comes from f (4) and f (5). . For example. with
f (2) = e−rδt q2 f (5) + (1 − q2 )f (4) . . a twostep tree is simply three branches. q3 ). But this initial fork from node 1 to nodes 2 and 3 also q2 = s2 exp(rδt) − s4 . (1 − q1 )(1 − q2 ) corresponding to the ‘probability tree’ (q1 . so for instance
s3 exp(rδt) − s6 . This is a case of the more general slogan that when working with independent events. must be the product of these.4. Expectation on a tree The expectation of some claim on the ﬁnal nodes of a tree is the sum over those nodes of the claim value weighted by the probabilities of paths reaching it. f (4). thus the pathprobability.
and similarly. the chance of going up and then down is q1 (1 − q3 ). under the probabilities q1 q3 . A twostep tree has four possible paths to the end. For claim pricing and expectation. the probability of following any particular path. but it is clear what it must be. But examine the expression we have derived above — it is of course precisely the expectation of the claims f (7).
Here qj is the probability sj exp(rδt) − s2j /(s2j+1 − s2j ). This means that f (3) comes from f (6) and f (7) via
f (3) = e−rδt q3 f (7) + (1 − q3 )f (6) . and f (2) if it was down. . discounted by the appropriate interestrate factor e−2rδt . it is worth f (3) if the ﬁrst jump was up. . q1 (1 − q3 ). one for the ﬁrst time step and one for the second. the chance of going up twice is the product q1 q3 . s7 − s6 But now we have a value for the claim at time 1. s5 − s4
and
q3 =
has the single branch structure.
We haven’t formally deﬁned expectation on our tree. the probabilities multiply.
.16
CHAPTER 2. in ﬁgure 2.
Thus the value of the claim at time zero has the daunting looking expression formed by combining the three equations above. And so on. But each path carries two probabilities attached to it. DISCRETE PROCESSES
onestep branch. (1 − q1 )q2 . and so on. The expectation of a claim is then the total of the four outcomes each weighted by this pathprobability.
f (1) = e−2rδt q1 q3 f (7) + q1 (1 − q3 )f (6) + (1 − q1 )q2 f (5) + (1 − q1 )(1 − q2 )f (4) . q2 . Its value at time zero must be
f (1) = e−rδt q1 f (3) + (1 − q1 )f (2) . Path probabilities The probability that the process follows a particular path through the tree is just the product of the probabilities of each branch taken.
All nodes here have claim values and are in pairs. This is the timezero value of the ﬁnal derivative claim — why? Because just as for the single branch.) Thus the nodes at time (n − 1) are all roots of branches that end on the claim layer and have arbitrage guaranteed values for the derivative attached — claimvalues in their own right now insisted on not by the investor’s contract (that only applies to the ﬁnal layer) but by arbitrage considerations. Consider any one of these ﬁnal branchings. there is a construction portfolio which. The fact that the tree is simply lots of branches was enough to banish the strong law here as well.2. Perverse though it may seem for a guaranteed construction procedure. And as fate casts the die and the stock jumps on the tree. Each layer will ﬁx the value of the derivative on the layer before. We will reach the root of the entire tree with a single value. We now have some idea of the complexity of the construction portfolios that will be required. This is the inductive step — we have moved the claims on the ﬁnal layer back one step. just like the stock. The investor ﬁlled in the nodes at the end of the tree with claims — we ﬁlled in the rest by constructing (φ. so this amount will jump as well. because each layer is only separated from the layer before by simple branches. the ends of single branchings. The results from section 2.
.1 provide a riskfree construction portfolio (φ. All claims can be constructed from a stock and bond portfolio. THE BINOMIAL TREE MODEL
17
The inductive step
Returning to our general tree over n periods. we now have a whole number of them. Instead of a single amount of stock φ. we will sweep backwards through the tree. (Both our grownup stock and the cash bond are indistinguishable over a single branching from the stock and bond of the simple model. will inexorably lead us to the claim payoff required. ψi ) are also random. ψ) of stock and bond at the root of the branch that can generate the time n claim amount. Arbitrage has worked its way into the tree model as well. What we have done is essentially a recursive ﬁlling in process. from a node at time (n − 1) to two nodes at time n. unlike the stock value they are known onestep in advance. and thus all claims have an arbitrage price. whatever path the stock actually takes.2. we start at its ﬁnal layer. Thus we can work back from enforced claims at the ﬁnal layer to equally strongly enforced claimvalues at the layer before.
The inductive result
By repeating the inductive step. one per node. But there is a vital structural difference — they are known just in time to be useful. though it will change at each tick time. the construction portfolios (φi . ψ) portfolios at each branching which guaranteed the correct outcome at the next step.
)
Figure 2.
A worked example
We can give a concrete demonstration of how this works. The tree nodes are the stock prices. now is
fnow = qfup + (1 − q)fdown .5: A stock price on a recombinant tree
What is the value of an option to buy the stock for 100 at time 3? It is easy to ﬁll in the value of the claim on the time 3 column. here we will see that the expectation operator with respect to some suitable set of ‘probabilities’ also provides the correct global structure for a hedge. s. these equations are a little simpler. As the interest rate r is zero. but the reemergence of the expectation operator is not just a coincidence peculiar to the simplicities of the branch model. Such trees are computationally much easier to work with. f . Yet again the expectation operator will appear with the correct result — just as the conclusion from the previous section was that with respect to a suitable ‘probability’. but what about expectation? We had no need of the probabilities pj .
. as long as we remember that there is more than one path to the ﬁnal nodes. (For simplicity. at the same node. the expectation operator provided the correct local hedge. Reading from top to bottom.5 is called recombinant as different branches can come back together. then the (riskneutral) probability q is
q= snow − sdown sup − sdown
and the value of a claim. If we are about to move either ‘up’ or ‘down’. interest rates are zero. DISCRETE PROCESSES
Expectation again
The strong law may be useless. 20. The tree in Figure 2. the claim has values then of 60. or recombine. 0 and 0. and at each node the process will go up with probability 3/4 and down with probability 1/4. We shall now need our equations for the new probabilities q and the claim values f .18
CHAPTER 2.
we should hedge
φ= fup − fdown sup − sdown
units of stock. At the end of this process we have the completed tree (ﬁgure 2.7).6 shows the result of the ﬁrst two such calculations. The price of the option at time zero is 15. We can complete ﬁlling in the nodes on level 2. We can trace through our hedge.6: The option claims and claimvalues at time 2
Figure 2. at any current time.2. and then repeat the process on level 1. using the formula that.2. THE BINOMIAL TREE MODEL
19
We calculate that the new qprobabilities are exactly 1/2 at each and every node.
Figure 2.7: The option claim tree
. Figure 2. and so on. Now we can work out the value of the option at the penultimate time 2 by applying the updown formulae to the ﬁnal nodes in adjacent pairs.
We buy an extra 0.25.50 0.5 units of stock costs 50. What would have happened if the initial jump had been down? Suppose the stock goes down to 80 Time 1 This time. V. Suppose the stock goes up again to 100 Time 2 The next hedge is (20 − 0)/(120 − 80) = 0. The option value matches the worth of both the old and the new portfolios. φ and ψ) depend on the sequence of updown jumps.
Table 2. Suppose the stock goes up again to 140 Time 2 The new φ is (60 − 20)/(160 − 120) = 1.00 Bond Holding ψi — 35 65 100
This was the rosy scenario. We calculate φ as (25 − 5)/(120 − 80) = 0. φ and ψ are random too. exactly canceling the debt. so overall we have broken even.25 units of stock and our borrowing mounts to 40. Finally suppose the stock goes down to 120 Time 3 The option will be in the money.) The table below shows exactly how the various processes change over time. We note that all the process above (S. Suppose the stock now goes up to 120 Time 1 The new φ is (40 − 10)/(140 − 100) = 0. Suppose the stock goes down again to 80 Time 3 Our stock is now worth 40.75 1. and we are exactly placed to hand over one unit of stock and receive 100 in cash to cancel our debt.5.25 units of stock at its new price. Buying 0. so we buy another 0. for instance V1 equals both φ1 S1 + ψ1 and φ2 S1 + ψ2 . φ1 units of stock are held during the interval from i = 0 to i = 1. the new φ is (10 − 0)/(100 − 60) = 0. The portfolio strategies shown are those in force for the previous tickperiod. so we take our stock holding up to 1. for instance.1: Option and portfolio development Stock Time i 0 1 2 3 Last Jump — up up down Price Si 100 120 140 120 Option Value Vi 15 25 40 20 Stock Holding φi — 0.50.75. (In fact. In particular.20
CHAPTER 2. taking our total borrowing to 65. DISCRETE PROCESSES
Time 0 We are given 15 for the option. making our debt now 100. the same would have happened if the stock had gone up to 160 instead. We sell half our stock holding and reduce our debt to 15. so we need to borrow an additional 35. but depend only on the jumps
. But the option is out of the money.
the chances of each of the ﬁnal nodes are (running from top to bottom) 1/8. B0 = 1 measure made up of the qs time of claim payoff
%
X : claim payoff
δt: length of period
&
.2. THE BINOMIAL TREE MODEL
21
Table 2.
xxxx.25 0. and a claim expectation of 33. The tree structure ensured that any claim provides just one possible value for its implied derivative instrument at every node or else arbitrage intervened. TNxxx
Exercise 2. The expectation result is still here.2. (That gives node probabilities of 27/64.50 0.50 Bond Holding ψi — 35 15 40
made up to the time when you need to work them out. Arbitrage spreads into every branch and thus across any tree.)
Conclusions
We can sum up. Claim led to claimvalue led to claimvalue via backwards induction until the entire tree was ﬁlled in. 27/64. Under the probability q . 3/8.2: Option and portfolio development along a different path Stock Time i 0 1 2 3 Last Jump — down up down Price Si 100 80 100 80 Option Value Vi 15 5 10 0 Stock Holding φi — 0.75. 3/8. The expectation of the claim is indeed 15 under these probabilities. and 1/8. 9/64 and 1/64. but certainly not under the model probabilities of 3/4up and 1/4down.2 Repeat the above calculations for a digital contract which pays off 100 if the stock ends higher than it started.
' $
Summary
q= erδt snow − sdown sup − sdown
−1 EQ BT X
φ=
fnow = e−rδt qfup + (1 − q) fdown V = f (1) = q: f: φ: ψ: V:
fup − fdown sup − sdown
−1 ψ = Bnow (fnow − φsnow )
arbitrage probability of upjump claim value timeprocess stock holding strategy bond holding strategy claim value at time zero
r: s: B: Q: T:
interest rate in force over period stock price process bond price process.
3
Binomial representation theorem
The expectation operator is much more general and constructive an operator than its conventional probabilistic role suggests. It is in this spirit.9: Tree with price process
(i) We will call the set of possible stock values. But a string of local construction portfolios is a global construction strategy guaranteeing a value.22
CHAPTER 2. There are seven separate deﬁnitions and each will be illustrated by an example on the double forked tree with seven nodes (ﬁgure 2. The cost of the local construction portfolio (φj . One possible process S on our tree is shown in ﬁgure 2.8).8: Tree with node numbers
Figure 2.
2. one for each node of the tree. S1 is either 60 or 120 depending on whether we are at node 2 or node 3. already met informally. We can raise the apparently coincidental ﬁnding that there exists some set of qj under which any derivative can be priced by a numerically trivial discounted expectation operation to the status of a theorem.
Figure 2.
.
Illustrated deﬁnitions
We must start with some formal deﬁnitions of concepts we have. ψj ) can be written as a discounted expectation. in many cases. DISCRETE PROCESSES
Something else happened as well — each branchlet carries its own probability qj under which ﬁxing the value at the branchlet’s root can be given by a local expectation operator with parameter qj . for instance. The random variable Si denotes the value of the process at time i. when we move to continuous models we shall be glad of any guidance — in the continuous case intuition will often fail. Though it seems strangely formal here where we have the comfort of a pictured tree. then. And far from vanishing. a process S . Thus the global discounted expectation operator gives the value of claims on a tree as well.9. that we derive the binomial representation theorem. and their pattern of interconnections. the expectation result carries across to the continuous model with ease.
6} 7 {1. 5} 6 {1. 7}
at time i. 3. we will call the set of ‘probabilities’ (pj ) or (qj ) a measure P or Q on the tree. and thus ﬁxes a node. the strong law failed precisely because it paid attention to both S and P.2. In full the ﬁltration associated with each node is It thus corresponds to a particular node achieved
Table 2. the probability of moving upwards from node j .
. That measure was a function of S and no function of P. represented by pj . the ﬁltration will either be F1 = {1. 2. To know where you are is the same as knowing the ﬁltration (at least in nonrecombinant trees). 3} if it was up. not S alone. The measure describes how likely any up/down jump is at each node. We could choose a simple measure P with all jumps equally likely (ﬁgure 2. 2} 3 {1. (iii) A ﬁltration (Fi ) is the history of the stock up until ticktime i on the tree. that is F0 = {1}.10b). 3} 4 {1. The ﬁltration ﬁxes a history of choices. 2. By time 1. The ﬁltration starts at time zero with F0 equal to the path consisting of the single node 1. 3.3: The ﬁltration process node ﬁltration 1 {1} 2 {1. The size and interrelation of upmoves affects the values of derivatives. the probabilities of achieving them does not. or F1 = {1.3. We didn’t need the real world measure P in order to ﬁnd the measure which allowed riskfree construction. Notice that in our formal system. and where the upmove is to. Put crudely. BINOMIAL REPRESENTATION THEOREM
23
(a) The measure P
(b) The measure Q
Figure 2. This separation of process and measure isn’t artiﬁcial — it is fundamental to everything we have to do. we have separated two components that would normally be seen as intimately connected parts of the same whole — the probability of an upmove. Why? Because the binomial structure ensures it — check for yourself that there is only one path to any given node.10: Measures P and Q
(ii) Separate from the process S . 2} if the ﬁrst jump was down.10a) or a more complex measure like Q (shown in ﬁgure 2. 4} 5 {1. They may not seem too different in character but the lesson of both the preceding sections is that this intuitive elision is unwise.
is a claim.6} {1.7} {1. we take P to be the measure in ﬁgure 2. thanks to the onetoone relationship between nodes and paths. For instance.3.
Table 2. We regard the node reached at time i as the new root of our tree.2. S2 } 180 120 80 80
The crucial difference between a claim and a process. S1 .
Table 2. as is the value of a call struck at 70 and the maximum price the stock attained along its path (table 2. and so is itself a random variable.5} {1.5: Conditional expectation against ﬁltration value Expectation EP (S2 F0 ) EP (S2 F1 ) EP (S2 F2 ) Filtration value {1} {1.3} {1.2} {1. EQ (XFi ) is the expectation of X if we have already got to that node. For each node at time i.
.4: Some claims at time 2 time 2 node 7 6 5 4 S2 180 80 72 36 (S2 − 70)+ 110 10 2 0 max{S0 .3. and it is useful to take expectations from later starting points. As an example. Or equivalently it is a function of the ﬁltration FT . the value of the process at time 2. The measure Q we might have guessed — it tells us which ‘probabilities’ to use in determining pathprobability and thus the expectation.10a and X to be the claim S2 (table 2. But so far we have only been interested in taking expectations along the whole of a path from time zero.4). while a process is deﬁned at all times up to and including T . the quantity EQ (XFi ) is the expectation of X along the latter portion of paths which have initial segment Fi .24
CHAPTER 2. For a claim X . is that the claim is only deﬁned on the nodes at time T . This conditional expectation has an enforced dependence on the value of the ﬁltration Fi .5).2. S2 . The ﬁltration serves this purpose. DISCRETE PROCESSES
(iv) A claim X on the tree is a function of the nodes at a claim timehorizon T .4}
1 2 (180 1 2 (72
Value (180 + 80 + 72 + 36)/4 = 92 + 80) = 130 + 36) = 54 180 80 72 36
(v) The conditional expectation operator EQ (·Fi ) extends our idea of expectation to two parameters — a measure Q and a history Fi . and take expectations of future claims from there.
starting at the root gives the same answer as the unconditioned expectation EP (S2 ). We could also see EP (XFi ) as a process in i. for every possible value of the ﬁltration F2 . What can we say about a previsible process? Given the onetoone relationship between nodes and histories on our binary tree. as is the delayed price process φi = Si−1 . whereas ‘starting’ at time 2 leaves no further time for development. it is certainly a binomial tree process in its own right.
Figure 2.
Figure 2.3. It is not always sensible to deﬁne the value that a previsible process has at time zero.12: The previsible process Si−1
Previsible processes will play the part of trading strategies.11. Fi−1 . But compared to the main process S . i ≥ 1 (ﬁgure 2. where we cannot tell in advance where prices are going to go. BINOMIAL REPRESENTATION THEOREM
25
Sensibly enough. In this way we can convert a claim into a process. it is shown in ﬁgure 2. It doesn’t seem to notice branches until one timestep after they have happened.2. so EP (S2 F2 ) = S2 . given a measure. This is an essential feature of any
. whose values are well deﬁned at each node later than time zero. it is known one node in advance. For instance a random bond price process Bi would be previsible.11: Conditional expectation process EP (S2 Fi )
(vi) A previsible process φ = φi is a process on the same tree whose value at any given node at timetick i is dependent only on the history up to one timetick earlier.12). In the case of X = S2 .
an easy check of whether a process is a Pmartingale or not is to compare the process Si itself with the conditional expectation process of its terminal value EP (ST Fi ). Rewritten again. For instance EQ (S1 F0 ) equals 1 × 120 + 2 × 3 3 60 = 80. then its conditional expected value under P is 100 thereafter. Example (3). that means the process S has no drift under P. 2 The last example above is a particular example of a general result. or is it special in some other way as well? (vii) A process S is a martingale with respect to a measure P and a ﬁltration (Fi ) if
EP (Sj Fi ) = Si .10b. Example (1). that conditioning ﬁrstly on the history up to time j and then conditioning on the history up to an earlier time i is the same as just conditioning originally up to time i.
In other words. for S to be a martingale with respect to a measure P. Slightly harder. This result is called the tower law. The process which constantly takes a ﬁxed value is. which matches the value S1 takes if the ﬁrst jump is up. The conditional expectation process of a claim For any claim X .
for all i ≤ j. Example (2). no bias up or down in its value under the expectation operator EP . DISCRETE PROCESSES
model that excludes arbitrage (or insider trading).
. Our ﬁnal deﬁnition is probably the most important of all — one question that we must surely ask soon is: what is the riskfree construction measure? Is it speciﬁc to the task in hand. it has no meaning without one) conditional on its history up until time i is merely the process’ value at time i. Only if these are identical is the process a Pmartingale. Our illustrative process S is actually a martingale under the measure Q given in ﬁgure 2. Written out. The downjump case and all the others need to be checked separately. The conditional expectation process Ni = EP (S2 Fi ) is a Pmartingale. Because of the nature of its deﬁnition we only need to check that 1 EP (N1 F0 ) is equal to N1 . Given the tower law. If the process has value 100 at some point. a martingale with respect to all possible measures. it is immediate. we need to use the fact that
EP EP (XFj )Fi = EP (XFi ). it means that the future expected value at time j of the process S under measure P (for of course our formal expectation demands a measure.26
CHAPTER 2.
This daunting expression needs expansion. rather trivially. and 80 is indeed the value of S0 . As this is just 1 × 130 + 2 × 54 = 92. the process EP (XFi ) is always a Pmartingale. To see this to be true. EQ (S2 F1 ) equals 2 3 5 × 180 + 5 × 80 = 120 if the ﬁrst jump was up. i ≤ j.
The structure of the tree ensures that the history Fi has two choices beyond Fi−1 .3 Check that EQ (S2 Fi ) is the same as Si .
.
where ∆Si := Si − Si−1 is the change in S from ticktime i − 1 to i. it is a Pmartingale. but it is a Qmartingale. The increments over the branch of the processes S and N are
∆Si = Si − Si−1
and
∆Ni = Ni − Ni−1 . Such a Q is called a martingale measure for S .3. We can get from N0 to Ni previsibly. If N is any other Qmartingale. And of course.13: The branch geometry (process S on left. TNxxx
Exercise 2. this kind of manipulation should be second nature.
Figure 2. our illustrative process S is not a Pmartingale (because ﬁgure 2. where Q is given in ﬁgure 2. and φi is the value of φ at the appropriate node at ticktime i. exactly the same process can be a martingale with respect to one measure and not to another. process N on right)
Consider a single branching from a node at ticktime i − 1 to two nodes ‘up’ and ‘down’ at ticktime i. For instance. then there exists a previsible process φ such that
i
Ni = N0 +
k=1
φk ∆Sk . BINOMIAL REPRESENTATION THEOREM
27
We must also take the P dependence seriously. and so prove that S is a Qmartingale. The proof is formal but straightforward — with the work we have put in already. corresponding to the up jump and down jumps respectively.
Binomial representation theorem
We can now write down our theorem. The process S is not a martingale on its own. with steps we know in advance.
xxxx. Binomial representation theorem Suppose the measure Q is such that the binomial price process S is a Qmartingale. it is a martingale with respect to the measure P.9 and ﬁgure 2.11 are different).2.10b.
so any random variable dependent on the branch is fully determined by its width size and a constant offset depending only on Fi−1 . Nowhere in our proof do we consider portfolios of a stock and bond. so we also have EQ (φi ∆Si Fi−1 ) = 0. wherever that might be.2. the stock follows a binomial process S . The scaling factor is previsible.
Financial application
We now have a theorem. it will in general be a construction based on a scaling (to match the widths) and a shift (to match the offsets). So we deﬁne φi to be the ratio of these branch widths. but it is a formal theorem about binomial tree processes and measures. starting where it starts and ﬁnishing where it ends. this k again determined only by Fi−1 . There are only two places to go. for φi and k known by Fi−1 . that is
φi = δni . So if we want to construct one random process out of another. both of these dependent only on the ﬁltration Fi−1 . That is ∆Ni = φi ∆Si + k. δsi
Now we can worry about the shift — the N increment ∆Ni must be given by the scaled increment φi ∆Si plus an offset k . And if there were a measure Q which made S a martingale. DISCRETE PROCESSES
The variability that these increments contain depends on the geometry of the branch itself (ﬁgure 2. At each point we could buy the appropriate φi of the stock and we would follow the gains and losses of the martingale Ni .28
CHAPTER 2.
And of course induction ties all these increments together to give the result we want. We would be able to match the martingale step for step. So the general scale and shift reduces in the case where S and N are both Qmartingales to just a scaling
∆Ni = φi ∆Si . that is EQ (∆Ni Fi−1 ) and EQ (∆Si Fi−1 ) are both zero — the increments have zero expectation conditional on the history Fi−1 . The previsible φ from the theorem could act as a construction strategy. Consider then the scaling ﬁrst. than that claim would have been synthesized. We go through many of the same steps as we had to in section 2. that is known by time i − 1. Thus the offset k must be zero as well (0 = 0 + k). nowhere do we consider arbitrage or market implications. If the martingale ended in a claim. How then can we use the binomial representation theorem for pricing? In our binomial tree model for the market. The size of the difference between the up and down jump values is δsi = sup − sdown for S and δni = nup − ndown for N .13). we could use the representation theorem to represent some other martingale Ni in terms of the stock price.
. But S and N are Qmartingales. but we haven’t reached a ﬁnancial conclusion.
a measure under which Si is a Qmartingale. One dollar today is like Bi dollars at time i. though. we need a ψi as well. we have not just a stock but a cash bond as well. buy the portfolio Πi consisting of:
. But we would like to be in a world without the growth of money — so we could simply factor it away. the claim doesn’t start or end anywhere. By the binomial representation theorem. Given any measure Q. there is a previsible process φ such that
i
Ei = E0 +
k=1
φk ∆Zk . now?
Construction strategy
Let’s try out the trick. Call this the discount process. the appropriate Ei is one as well. (i) The process Bi−1 is another previsible process. all things being equal. The cash bond Bi represents the growth of money — $1 today is not the same as $1 at time i.
by taking conditional expectations. whatever measure Q we choose. X ray vision or intuition would suggest that the φi of the binomial representation theorem is going to be a vital part of our formal construction strategy but. (ii) Deﬁne Zi := Bi−1 Si which is just as well a deﬁned process as S itself and it subsists on the same binomial tree. BINOMIAL REPRESENTATION THEOREM
29
Two things stand in our way. it’s a random variable. Even better. −1 Ei = EQ (BT XFi ). Firstly we have a claim X . Call this the discounted claim. as we have already observed. The claim X is a random variable but we have already seen one trick for turning random variables into processes. we can form the process
Ei = EQ (XFi ).
Now consider the following construction strategy: at ticktime i. Secondly. to use the notation of earlier. With each stock holding comes a bond holding. First things ﬁrst. Ei is automatically a Qmartingale. The bond process Bi is previsible and positive. With Q such that Z is a Qmartingale and claim X .
−1 (iii) The value BT X is also a claim and because of the simple mapping from Z to S it’s as much a claim on Z as S . It isn’t a process.2. Thus if we ﬁnd Q. We can assume without loss of generality that B0 = 1. What about the cash bond? Ultimately we will simply have to grind through the algebra but a bit of intuition can guide us to what the answer might look like. And though we would like to end up at the claim. then. Call this the discounted stock process. just like Bi itself.
What. there −1 is a Qmartingale process produced from BT X by taking conditional expectations. not a martingale.3.
but its constituents have changed in value: Π0 is now worth −1 −1 φ1 S1 + ψ1 B1 = B1 E0 + φ1 (B1 S1 − B0 S0 ) . to be the opening value of the portfolio Πi at time i. What about one tick later? We have held the portfolio safe across the period. the claim we require. DISCRETE PROCESSES
• φi+1 units of the stock S . But the portfolio Π1 . ψi ) strategy which duplicates any claim. −1 −1 but B1 S1 − B0 S0 = ∆Z1 . Π0 is worth B1 E1 .
. In symbols. under the martingale measure Q for the discounted stock Z . And at the end of our selfﬁnancing strategy. which we need to create at time 1. which is BT BT X . We shouldn’t be too surprised — we are simply repeating the argument of section 2. Then a strategy is selfﬁnancing if the closing value of the portfolio Πi−1 at time i is precisely equal to Vi . that is whichever ﬁltration F1 actually obtains. There is also no difﬁculty in determining φ1 or ψ1 as φ and ψ are previsible. Our construction strategy is what we might call selfﬁnancing. costs precisely that amount above: B1 E1 .
Di = Vi − φi Si − ψi Bi . Π0 is worth φ1 S0 + ψ1 B0 = E0 = EQ (BT X) — it costs that much to create. whatever actually happened to S . the cost of the next portfolio. What do we mean exactly by selfﬁnancing? Let us deﬁne Vi . But our formal argument has won us an overview of the entire process and a couple of vital slogans:
The existence of selfﬁnancing strategies
The ﬁrst slogan is that within the binomial tree model. portfolio Πi costs Bi Ei to purchase. and it will change by time (i + 1) to be worth Bi+1 Ei+1 . Now we can use the binomial representation theorem to simplify the expression above: at time 1. And it is an arbitrage price because any other price could be milked for free money by running the (φi . At time i. the ‘ﬁnancing gap’ of cash that would otherwise have to be injected into the strategy.30
CHAPTER 2. ψi ) strategy the appropriate way round to duplicate the claim. −1 At time zero.
−1 • ψi+1 = (Ei − φi+1 Bi Si ) units of the cash bond. and the construction strategy demands that we buy a new portfolio Π1 .2 in formal guise. That is X . we can produce a selfﬁnancing (φi . we end up with the worth of −1 ΠT −1 at time T .
Arbitrage
−1 The price of the claim X is now obvious: it is EQ (BT X) — the expected value of the discounted claim. And so on. our starting point. that is Vi = φi+1 Si + ψi+1 Bi . We are at time 1. the worth of the trading strategy at time i.
must be zero. Thus we can cash in our portfolio Π0 to create Π1 .
2. (ii) the change in value V of the portfolio deﬁned by the strategy obeys the difference equation:
∆Vi = φi ∆Si + ψi ∆Bi
where ∆Si := Si − Si−1 is the change in S from ticktime i − 1 to i.
The gap Di at time i is zero if and only if the change in value of the strategy from time i − 1 to i is due only to changes in the stock and bond values alone.3. we can add almost as an afterthought that for any sensible stock process S . the discounted stock. Formally: Selfﬁnancing hedging strategies Given a binomial tree model of a market with a stock S and bond B .
Uniqueness and existence of Q
And in this discrete world.
∆Vi = φi ∆Si + ψi ∆Bi + Di . Option price formula (discrete case) The value at ticktime i of a claim X maturing at date T is
−1 Bi EQ (BT XFi ). (iii) and is identically equal to the claim X . justiﬁed by the binomial representation theorem. which requires that amount to start off and yields the claim without risk at T .
. BINOMIAL REPRESENTATION THEOREM
31
Another way of representing this selfﬁnancing property comes from the changes of the strategy value process ∆Vi = Vi − Vi−1 . then (φi .
Why? Precisely because there is a selfﬁnancing strategy.
Expectation of the discounted claim under the martingale measure
The second of these slogans is that the price of any derivative within the binomial tree model is the expectation of the discounted claim under the measure Q which makes the discounted stock a martingale. ψi ) is a selfﬁnancing strategy to construct a claim X if: (i) both φ and ψ are previsible. and ∆Bi := Bi − Bi−1 is the corresponding change in B . there will be a unique measure Q under which Bi−1 Si . is a Qmartingale.
32
CHAPTER 2. The real measure P which S follows is irrelevant. For a ﬁxed time t. As that quantity gets smaller. look into the continuous world with our discrete techniques. Any claim on a stock implies a derivative instrument tied to the underlying stock value at any time by a construction strategy capable of providing arbitrage riches if any market player disobeyed it. By the central limit theorem. The stock process follows the nodes of a recombinant tree. in a heuristic way. the model should ever more closely approximate a continuoustime model.
. we could believe that a continuous model can be approximated by a discrete time model with a very small intertick time. then n = t/δt and
√ St = S0 exp µt + σ t 2Xn − n √ n . and the riskless interest rate r. we have the general theorem we require.
2. if down. which does not depend on the interval size.4 Overture to continuous models
We can.
The jumps are all equally likely to be up as down. It will even be possible to ‘derive’ the BlackScholes option pricing formula. the stock growth rate µ. √ so that (2Xn − n)/ n has mean zero and variance one. Indeed we can show that a natural discrete model with constant growth rate and noise approximates a lognormal distribution under both the original measure P and the martingale measure Q. if we set n to be the number of ticks till time t.
Model with constant stock growth and noise
The model is parameterized by the intertick time δt. The construction strategy is selfﬁnancing and generates the claim whatever S does. There are also three ﬁxed and constant parameters: the noisiness σ . the measure Q under which the discounted stock is a martingale. but expectation under just one special measure. The random variable Xn has the binomial distribution with mean n/2 and variance n/4. DISCRETE PROCESSES
Conclusions
We are now ﬁnished in the discrete world. Without being fully rigorous yet. which moves from value s at some particular node along the next up/down branch to the new value
√ s exp µδt + σ δt s exp µδt − σ √δt
if up. The cash bond Bt has the simple form that Bt = exp(rt). though its rigorous development must wait until the very end of the next chapter. that is p = 1/2 everywhere.
where Xn is the total number of the n separate jumps which were upjumps. That arbitrage justiﬁed value is the expectation of the discounted claim.
√ √ 1 Thus (2Xn − n)/ n has mean − t(µ + 2 σ 2 − r)/σ and variance asymptotically approaching one. OVERTURE TO CONTINUOUS MODELS
33
this distribution converges to that of a normal random variable with zero mean and unit variance.
where Φ is the normal distribution function Φ(x) = Q(Z ≤ x). struck at k .
We will see in chapter three that this evaluates as
S0 Φ log S0 + r + 1 σ 2 T k √ 2 σ T − ke−rT Φ log S0 + r − 1 σ 2 T k √ 2 σ T . then its worth at time zero is
−1 EQ (BT X) = EQ
√ S0 exp(σ T Z − 1 σ 2 T ) − k exp(−rT ) 2
+
. The corresponding St is still lognormally distributed with log St having mean log S0 +(r− 1 σ 2 )t 2 and variance σ 2 t.4. We have found the marginal distribution of St under the martingale measure Q. This can be written
√ St = S0 exp σ tZ + (r − 1 σ 2 )t . under Q. with X = (ST − k)+ . Again the central limit theorem gives the convergence of this to a normal random variable with the same mean and variance exactly one. 1) under Q.
Under the martingale measure
This is what happens under the original measure P. Xn is still binomially distributed. sup − sdown
We can calculate that q is approximately equal to
1 q= 2 1− √ δt µ + 1 σ2 − r 2 σ . So as δt gets smaller and n gets larger.2. but what goes on with Q? Following our formula the martingale measure probability q is
q= s exp(rδt) − sdown . 2
where Z is a normal N (0. as log St is normally distributed with mean log S0 + µt and variance σ 2 t.
So. This is a preview of the BlackScholes formula which we shall prove properly in the next chapter. the distribution of St becomes lognormal.
. but now has mean nq and variance nq(1 − q).
Pricing a call option
If X is the call option maturing at date T .
With our discrete stock price model we didn’t have any old random process. We forcibly limited ourselves to a binomial tree. But such trees grow too complex and we stop being able to see the wood. Firstly. What is a continuous process? Three smallscale principles guide us. We shall have to start from scratch in the continuous world. The binary choice of a single jump ‘up’ or ‘down’ only becomes subtle as the ticks get closer and closer. the value can change at any time and from moment to moment. In practice. For the continuous world we need an analogous basis — something simple and yet a reasonable starting point for realism. We started simply and hoped (with some justiﬁcation) that complex enough market models could be built from such humble materials. If there is one overarching principle to this chapter it is that Brownian motion is sophisticated enough to produce interesting models and simple enough to be tractable. The discrete trees of the previous chapter are only an approximation to the way that prices actually move. the actual values taken can be expressed in arbitrarily ﬁne fractions — any real number can be
34
.
S
3. We will no longer be able to proceed in full generality — we will concentrate on Brownian motion and its relatives. rather than just at some ﬁxed ticktimes when a portfolio can be calmly rebalanced. Given the subtleties of working with continuous processes. The discrete models will guide us — the intuitions gained there will be more than useful — but limiting arguments based on letting δt tend to zero are too dangerous to be used rigorously.1 Continuous processes
We want randomness. giving the tree more and evershorter branches. The single binomial branching was the building block for our ‘realistic’ market. We will encounter a representation theorem which establishes the basis of riskfree construction and again it will be martingale measures that prime the expectation operator correctly. Secondly. But processes and measures will be harder to separate intuitively — we will need a calculus to help us. And changes in measure will affect processes in surprising ways.Chapter 3
Continuous processes
tock prices are not trees. a price can change at any instant. a simple calculus based on Brownian motion will be more than enough for us.
the global structure of the stock index is different. CONTINUOUS PROCESSES
35
taken as a value. And lastly the process changes continuously — the value cannot make instantaneous jumps. n n
for all i ≥ 1. We shouldn’t run ahead of ourselves. Even though they move in a ‘sharpedged’ way.3.05 it must have passed through. X2 . In other words. people have gone further and compared the prices to one particular continuous process — the process followed by a randomly moving gas particle. In other words. albeit quickly. . Brownian motion can’t be the whole story. Brownian motion will prove a remarkably effective component to build continuous processes with — locally Brownian motion looks realistic. gets ‘noisier’ as time passes. But we only want a basis — the single binomial branching didn’t look promising right away. and doesn’t go negative.1: UK FTA index. it isn’t too unrealistic to claim that they nonetheless display continuous process behavior. The ﬁrst step to the analysis of Brownian motion is to construct a special family of discrete binomial processes.2).1. the similarity fades — ﬁgure 3. 196392
Figure 3.
Figure 3.2: Brownian motion
Locally the likeness can be striking — both display the same jaggedness. and the same similarity under scale changes — the jaggedness never smooths out as the magniﬁcation increases. is a sequence of independent binomial random variables taking values +1 or −1 with equal probability. . or Brownian motion (ﬁgure 3. But globally. At least as a starting point. . then the value of Wn . at the ith step is deﬁned by:
Xi i Wn ( n ) = Wn ( i−1 ) + √ . And as far back as Bachelier in 1900. we can insist that stock market indices or prices of individual securities behave this way. if the value changes from 1 to 1. all the values in between.2.
Brownian motion
It was nearly a century after botanist Robert Brown ﬁrst observed microscopic particles zigzagging under the continuous buffeting of a gas that the mathematical model for their movements was properly developed. if X1 .
. who analyzed the motion of the Paris stock exchange. At an intuitive level.1 doesn’t look like ﬁgure 3. It grows.
But the distribution always has zero mean and unit variance. And so the distribution of Wn (t) tends to a normal
. to a normal N (0.)
Figure 3.36
CHAPTER 3. In fact. 1). the value of Wn (t) is the same as
Wn (t) = √ t
nt i=1 Xi √
nt
. 1) random variable. each with zero mean and variance 1/n.4) appear to be √ settling down towards something as n increases.3: The ﬁrst two steps of the random walk Wn
Figure 3. Can we make a formal statement? Consider for example. the distribution of Wn (1) tends towards the unit normal N (0. The moves of size 1/ n seem to force some kind of convergence.
The distribution of the ratio in brackets tends. by the central limit theorem. 64. What does Wn look like as n gets large? Instead of blowing out of control. the family portraits (ﬁgure 3.4: Random walks of 16. the distribution of Wn at time 1: for a particular Wn . CONTINUOUS PROCESSES
The ﬁrst two steps are shown in ﬁgure 3. ranging from − n to n. there are n + 1 possible √ √ values that it can take. 256 and 1024 steps respectively
Moreover the central limit theorem gives us a limit for these binomial distributions — as n gets large.3. (Because Wn (1) is the sum of n IID random variables.
t).1. In the continuous world. but all the conditional marginal distributions as well. CONTINUOUS PROCESSES
37
N (0. but all the marginals conditional on all the histories Fs . just as it was in the discrete. Each random walk Wn has the property that its future movements away from a particular position are independent of where that position is (and indeed independent of its entire history of movements up to that time).1 If Z is a normal N (0. Additionally such a future displacement Wn (s + t) − Wn (s) is binomially distributed with zero mean and variance t. then the process Xt = tZ is continuous and is marginally distributed as a normal N (0. though an exact echo of the behavior of the discrete precursors Wn (t). under P.
xxxx. t) are not Brownian motion. and W0 = 0. It will in fact be the daunting task of specifying all these that drives us to a Brownian calculus. the conditional marginals converge. and is independent of Fs . t). And indeed they do. it is not just the marginals (conditional on the process’ value at time zero) that count. and it converges towards Brownian motion.3. t). and all conditional marginals tend towards a normal distribution of the same mean and variance. is subtle. though this isn’t the place to set up the careful formal framework to make sense of that statement. Thus again. (ii) the value of Wt is distributed. Formally: Brownian motion The process W = (Wt : t ≥ 0) is a PBrownian motion if and only if (i) Wt is continuous. Is X a Brownian motion?
√
.
There is a formal unity underlying the family — all the marginal distributions tend towards the same underlying normal structure. t). And not just all the marginal distributions. The marginals converge. (iii) the increment Ws+t − Ws . under P. as a normal random variable N (0. TNxxx
Exercise 3. Many processes that have marginals N (0. 1). and the temptation is irresistible to say that the distributions of the processes converge too. the central limit theorem gives us a constant limiting structure. is distributed as a normal N (0. The last condition. These are both necessary and sufﬁcient conditions for the process W to be Brownian motion. The distribution of Wn converges. the history of what the process did up to time s.
Brownian motion is a fractal. And if it looks too noisy. It may be a million units above the axis.3 Show that. TNxxx
Exercise 3.
and then again from time to time in the future. σ = 91. But we can add in a drift artiﬁcially.
xxxx. whereas the stock of a company normally grows at some rate — and historically we expect prices to rise if only because of inﬂation.
Brownian motion as stock model
We had our misgivings about Brownian motion as a global model for stock behavior. for a constant noise factor σ . It has mean zero. for all values of σ (σ = 0). it immediately hits it again inﬁnitely often. and T > 0 there is always a positive probability that ST is negative. For example the process St = Wt +µt. we can scale the Brownian motion by some factor: for example. But in this particular case. and is a (onedimensional) Gaussian process. but it will (with probability one) be back down again to zero. How are we doing? Consider the stock market data shown in ﬁgure 3. or how negative.1. for some constant µ reﬂecting nominal growth.81] and simulate a sample path.2 If Wt and Wt are two independent Brownian motions and ρ ˜ is a constant between −1 and 1. (Hint: consider the marginal distribution of ST . it is (with probability one) differentiable
nowhere. The process went negative. We could estimate σ and µ for the best ﬁt [in this case.
• Once Brownian motion hits a value. Brownian motion wanders. St = σWt + µt. but we don’t have to use it on its own. as we want. CONTINUOUS PROCESSES
xxxx. µ.
• It doesn’t matter what scale you examine Brownian motion on — it looks just
the same. we have a glitch right away. Is this X a Brownian motion?
It is also worth noting just how odd Brownian motion really is.38
CHAPTER 3.
• Brownian motion will eventually hit any and every real value no matter how
large. t). by some later time. or not noisy enough. is called Brownian motion with drift. but here is a brief peek into the bestiary:
• Although W is continuous everywhere.)
. then the process Xt = ρWt + 1 − ρ2 Wt is continuous and has marginal distributions N (0. which we may not want for the price of a stock of a limited liability company. Not bad — the process has longterm upwards growth. We won’t stop to prove them. Brownian motion is often also called a Wiener process.3 and µ = 37. TNxxx
˜ Exercise 3.
3. pinning down a small section under a microscope.7%] we can simulate a sample path (ﬁgure 3.7.087.
3. In ﬁgure 3. not surprisingly well known and it is usually called exponential Brownian motion with drift.178 and µ = 0. Consider any smooth (differentiable) curve. 196392
Figure 3. Again ﬁnding a best ﬁt for σ and µ [σ = 0.1: UK FTA index. STOCHASTIC CALCULUS
39
Figure 3.
Figure 3.8% and an annual drift of 8. It is not the only model for stocks — and indeed we will look at others later on — but it is simple and not that bad.2.
. Globally it can have almost any behavior it likes. because the condition that it is differentiable does nothing to affect it at a large scale. or sometimes geometric Brownian motion with drift. a ‘noisiness’ of 17.6). increasing the magniﬁcation by a factor of about ten each time.5: Brownian motion plus drift
We can though be more adventurous in shaping Brownian motion to our ends. taking the exponential of our process:
Xt = exp(σWt + µt). (Could you tell which picture was which without the captions?) Brownian motion can prove an effective building block.2 Stochastic calculus
Shaping Brownian motion with functions may be powerful. Consider for example. we focus in on the point of a particular differentiable curve with xcoordinate of 1. but it brings a dangerous complexity.6: Exponential Brownian motion
This process is.
Now we mirror the stock market’s longterm exponential growth (and for good measure we start off quietly and get noisier). Suppose we zoom in though.7.
Consider. for example: (1) The equation dft = µdt. the slope or drift of the magniﬁed straight line at t. Newtonian calculus is the formal acknowledgement of this. we could decide to build up a family of nice functions by specifying how they are locally built up out of our building block. (2) The equation dft = tdt. If f0 . With a Newtonian construction. CONTINUOUS PROCESSES
Reading the graphs from left to right and line by line.
where µt is our scaling function. for some constant µ. If f0 = 0.
.
Figure 3.7
Differentiable functions. The going was a bit harder here. Here we have a slope at time t of value t — what does this look like? Simple integration comes to the rescue.40
CHAPTER 3. each small box is expanded to form the frame for the next graph. was equal to zero then we might guess (correctly) that ft could be written in more familiar notation as ft = µt. but we managed 2 it. however strange their global behavior. and we can check it ourselves by differentiation: ft = t as we require. We would write the change in value of a Newtonian function f over a time interval at t of inﬁnitesimal length dt as
dft = µt dt. the graph section becomes smoother and straighter. What is f ? That is. As the process continues. are at heart built from straight line segments. until eventually it is straight — it is a small straight line. for example. what does it look like? How does it behave globally? Could we draw it? If we stick together straight line segments of slope µ.7: Progressive magniﬁcation around the point 1. Then we could explore our universe of Newtonian functions. then we could again pin down ft as ft = 1 t2 . the straight line. then intuitively we just produce a straight line of slope µ.
t)dt
is called an ordinary differential equation (ODE). for f0 = 0. Speciﬁcally. It is clear that though ODE s are powerful construction tools. consider a construction procedure based on Brownian motion.
Instead of just giving the drift µt directly. they are also dangerous ones. then
dft = µ(ft . our intuition dismissed the possibility of another solution.3. as direct integration is not a route to the solution. Now things are harder. namely ft = a exp(−1/t) for every possible value of an arbitrary constant a.
˜ ˜ • If ft and ft are two differentiable functions agreeing at 0 (f0 = f0 ) and they ˜ ˜ have identical drifts (dft = dft ). Uniqueness of Newtonian differentials Two complementary forms of uniqueness operate here. but what about here? The construction metaphor (dft = tdt tells us how to build ft . We could guess — say ft = et — and then check by differentiation. there are no solutions at all. There are plenty of ‘bad’ ODEs which we haven’t a clue how to explore.
Stochastic differentials
And if it was bad for Newtonian differentials. there is only one drift function µt t which satisﬁes ft = f0 + 0 µs ds (for all t).8)
.2. then the processes are equal: (ft = ft ) for all t. where solutions need neither exist nor be unique. Given f0 = 0. Zooming in on Brownian motion doesn’t produce a straight line. Perhaps our universe of Newtonian functions isn’t so benign. (4) The equation dft = ft t−2 dt. it forms a solution. t). there are an inﬁnite number of solutions. and thus given a starting place and a deterministic 1 building plan we ought to produce just one possible ft ) suggests that ft = 2 t2 is the unique solution and indeed we can formalize this. There are plenty of ODEs which have no solutions. we might have a problem where the drift itself depends on the current value of the function. This solution happens to be unique for f0 = 1. given a differentiable function ft . and plenty more which do not have unique solutions. However. where µ(x. • Secondly. (The uniqueness of the solution to an ODE cannot be deduced just from the uniqueness of Newtonian differentials in the box. This is an example of a bad case.) (3) The equation dft = ft dt. t) is a known function. If there is a differentiable function f which satisﬁes it (with given f0 ). So µ is unique given f . In other words. f is unique given the drift µt (and f0 ). (ﬁgure 3. if the drift µt equals µ(ft . STOCHASTIC CALCULUS
41
What about uniqueness though? In the ﬁrst example.
The selfsimilarity of Brownian motion means that each new graph is also a Brownian motion. like X and σ . But of course this selfsimilarity is ideal for a building block — we could build global Brownian motion out of lots of local Brownian motion segments.8: ‘Zooming in’ on Brownian motion
As before. each box is expanded by suitable horizontal and vertical scaling to frame the next graph. and in particular it excludes discontinuous cases such as Poisson processes. CONTINUOUS PROCESSES
Figure 3. so can the noisiness σt . The Brownian term of X can have a noise factor σt . ﬁnding this out entails ‘integrating’ stochastic differentials in some way. And of course. Such processes.42
CHAPTER 3. This deﬁnition of stochastic process (see box) is not universal. Nevertheless it will be quite adequate for all the models we will meet. based on the inﬁnitesimal increment of W which we will call dWt . But it can also be random and depend on values that X (or indeed W ) took up until t itself.
. We can. whose value at time t can depend on the history Ft . are called adapted to the ﬁltration F of the Brownian motion W . and so the inﬁnitesimal change of Xt is
dXt = σt dWt + µt dt. A stochastic process X will have both a Newtonian term based on dt and a Brownian term.
Stochastic processes
What does our universe look like? As with Newtonian differentials.
As in the Newtonian case. If we built using straight line segments (suitably scaled) too. but not the future. we could include Newtonian functions as well. the drift µt can depend on the time t. formally deﬁne what it is to be a (continuous) stochastic process. We call σt the volatility of the process X at time t and µt the drift of X at t. and just as noisy. though. And we could build general random processes from small segments of Brownian motion (suitably scaled).
2 where σ and µ are random F previsible processes such that 0t (σs + µs )ds is ﬁnite for all times t (with probability 1). And as it happens. perhaps in terms of Wt . if two processes Xt and Xt agree at time zero (X0 = X0 ) and they ˜ have identical volatility σt and drift µt . the Brownian motion we have some handle on? Partially.2. and that they may have some jump discontinuities. then the processes are equal: Xt = Xt for all t. the equation
dXt = σ(Xt . In other words. In the simple case. t) is some deterministic function. t)dt
is called a stochastic differential equation (SDE) for X . but also on other events in the history Ft .
Uniqueness of volatility and drift Two complementary forms of uniqueness operate here. and if it does it might not be unique. X is unique given σt and µt (and X0 ). meaning that X
. given the process X . This class is closed under all the operations used later. Usage of the term SDE does tend to spread out from this strict deﬁnition to include the stochastic differentials of processes whose volatility and drift depends not only on Xt and t. where σ and µ are both constants. such as σt = σ(Xt . Stochastic processes A stochastic process X is a continuous process (Xt : t ≥ 0) such that Xt can be written as
t t
Xt = X0 +
0
σs dWs +
0
µs ds. t). there is only one pair of volatility σt and drift t t µt which satisﬁes Xt = X0 + 0 σs dWs + 0 µs ds (for all t). an SDE need not have a solution. As in the Newtonian case (ODEs). The differential form of this equation can be written
dXt = σt dWt + µt dt. But can we recognize the world we have created. In terms of stochastic analysis.
˜ ˜ • Firstly. And it will generally be easier to write down the SDE (if it exists) for a particular X then it is to provide an explicit solution for the SDE. this deﬁnes stochastic processes to be semimartingales whose drift term is absolutely continuous. we can provide a uniqueness result to mirror the classical setup. • Secondly. where σ(x. In the special case when σ and µ depend on W only through Xt . t)dWt + µ(Xt . and all the models considered will lie within it. STOCHASTIC CALCULUS
43
The technical condition that σ and µ must be F previsible processes means that they are adapted to the ﬁltration F .3. This uniqueness of σt and µt given X comes from the DoobMeyer decomposition of semimartin
gales.
that increment is independent of the Brownian motion up to that point. and so on. so 2Wt dWt cannot be the differential of Wt2 . which means that so too must the product of the increment and W (it/n). that is
t n−1
2
0
Ws dWs ≈ 2
i=1
W
it n
W
(i+1)t n
−W
it n
. t/n. CONTINUOUS PROCESSES
has constant volatility and drift. Could we use a simple chain rule to produce the stochastic differential dft ? Under Newtonian rules. say f (Wt ) = Wt2 . because of the variance structure of Brownian motion. . because its integral doesn’t even have the right expectation. t] into a partition {0.
dXt = Xt (σdWt + µdt). the SDE for X is
dXt = σdWt + µdt. (n − 1)t/n.3 Itˆ calculus o
Intuitive integration doesn’t carry us very far. And our meager understanding of Wt and dWt at least gives us some conﬁdence that the differential form of σWt is σdWt . because
t
if
0
2 d(Ws ) = 2
t 0
Ws dWs . We need tools to manipulate the differential equations. Also the increment has zero mean.
Now something begins to worry us.
(assuming that X0 = 0). product rule. just as Newtonian calculus has the chain rule.44
CHAPTER 3.
It isn’t too hard to guess what the solution to this is:
Xt = σWt + µt.
How can we tackle 0t Ws dWs ? Consider dividing up the time interval [0. which doesn’t look too implausible. integration by parts. . How far could Newton carry us? Suppose we had some function f of Brownian motion. . The difference term inside the brackets is just the increment of Brownian motion from one particular partition point to the next. But consider the only slightly more complex SDE (echoing the Newtonian ODE of example (3) above). As σ and µ are independent of X .
. But Wt2 has mean t.
then
Wt2 = 2
t 0
Ws dWs .
We’re at sea. t} for some n. the uniqueness result could form a part of a proof that this is the only solution.
3. 2t/n. Then we could approximate the integral with a summation over this partition. By property (iii) of Brownian motion. So the summation consists of terms with zero mean. But we should check via integration. forcing it to have zero mean itself. dWt2 would be 2Wt dWt . and in particular it is independent of the Brownian motion term W (it/n). .
. we assumed that (dWt )2 and higher terms were zero. by Brownian motion fact (iii).i = W
ti n
−W t/n
t(i−1) n
. (Because each increment W ti − W t(i−1) a normal N (0.
then for each n. Thus 0t (dWs )2 = t. the sequence Zn. . t/n). which is negligible compared with dt.ˆ 3. it only looks second order because of the notation. . We can’t ignore (dWt )2 . then Yt := f (Xt ) is also a stochastic process and is given by
1 2 dYt = σt f (Xt ) dWt + µt f (Xt ) + 2 σt f (Xt ) dt. Take (dWt )2 . independent of the ones n n before it.3. and f is a deterministic twice continuously differentiable function. t] we just used: {0. the distribution of the right hand side summation 2 converges towards the constant expectation of each Zn. t/n. we can apply Itˆ with X = W and f (x) = x2 and we have o
d(Wt2 ) = 2Wt dWt + dt. 2t/n. satisfying dXt = σt dWt + µt dt. We can model the integral of (dWt )2 by the (hopefully convergent) approximation
t 0 n
(dWt ) =
i=1
2
W
ti n
−W
t(i−1) n
2
. . is a set of IID normals N (0.i
n
.
But if we let Zn.1 . But as we have observed before. o o Itˆ ’s formula o If X is a stochastic process. Zn.) We can rewrite our approximation for (dWs )2 as
t 0 n
(dWs ) ≈ t
i=1
2
2 Zn. What about (dWt )3 and so on? It turns out that they are zero. t}.i be
Zn.) So Taylor gives us:
1 df (Wt ) = f (Wt )dWt + 2 f (Wt )dt + 0.i .
or
Wt2 = 2
t 0
Ws dWs + t.
By the weak law of large numbers (just like the strong law but only talking about the distribution of random variables). ITO CALCULUS
45
What went wrong? Consider a Taylor expansion of f (Wt ) for some smooth f :
1 df (Wt ) = f (Wt )dWt + 2 f (Wt )(dWt )2 + 1 3! f
(Wt )(dWt )3 + · · ·
Overfamiliar with Newtonian differentials.2 .
. given the same partitioning of [0. Brownian motion is odd.
Returning to our Wt2 . or in differential form (dWt )2 = dt. E(dWt 3 ) has size (dt)3/2 .
The formal version of this surprising departure from Newtonian differentials is the deservedly famous Itˆ ’s formula (sometimes seen modestly as Itˆ ’s lemma). . namely 1. . . (For example. 1).
The exponential function is particularly pleasant. But of course the Xt we want can be written as Xt = f (Yt ). Or in other words. and f to be the exponential function f (x) = ex . TNxxx
Exercise 3.
xxxx. using Itˆ to convert o processes to SDEs is relatively straightforward.4 If Xt = exp(Wt ).
Here.
What SDE does X follow? We know we can handle the term inside the brackets but we have to take a stochastic differential of the exponential function as well. we can use Itˆ ’s formula.
Processes from SDEs
Much like differentiation (easy. More generally.46
CHAPTER 3. to solve them. Most stochastic differential equations are just too difﬁcult to solve. But a few. because it is the volatility of the process log Xt . as f (Yt ) = f (Yt ) = f (Yt ) = Xt . o Suppose we took Yt to be the process σWt + µt. We will also use the name logdrift for the drift µ of log Xt . which is different from the drift of dXt /Xt above. and which is often abbreviated just to volatility notwithstanding that term’s existing deﬁnition. then what is dXt ?
SDE s
from processes
Itˆ ’s most immediate use is to generate SDEs from a functional expression for a proo cess. With the right formulation though. but its inverse can be impossible). Consider the exponential Brownian motion we set up in section 3. In general we can’t. the variable σ is sometimes called the logvolatility of the process. And if that were all we ever wanted to do there would be few problems. then f (X) has differential
df (Wt ) = f (Wt )dWt + 1 f (Wt )dt. But it isn’t — one of the key needs we have is to go in the opposite direction and convert SDEs to processes. and just like some ODEs they depend on an
. Then Yt is simple enough that we can write down its differential immediately: dYt = σdWt + µdt. if X is still just the Brownian motion W . CONTINUOUS PROCESSES
which at least has the right expectation. rare examples can be. so we can rewrite the differential as
1 dXt = Xt σdWt + µ + 2 σ 2 dt . 2
as hinted above.1:
Xt = exp(σWt + µt). so one application of Itˆ ’s formula gives us o
1 dXt = σf (Yt )dWt + µf (Yt ) + 2 σ 2 f (Yt ) dt.
We guess then that
Xt = exp σWt − 1 σ 2 t . we can see that
d(Xt Yt ) = Xt dYt + Yt dXt + σt ρt dt.3. then the drift term in the SDE would match our SDE as well. Let’s go back then to the SDE we tripped over earlier:
dXt = Xt (σdWt + µdt). ITO CALCULUS
47
inspired guess and then a proof that the proposed solution is an actual solution via Itˆ .5 What is the solution of dXt = Xt (σdWt +µt dt).ˆ 3.
We need an inspired guess — so we notice that the stochastic term (σXt dWt ) from this SDE is the same as the SDE we generated via Itˆ in the section above.
We could match both drift and volatility terms for this SDE and the SDE of exp(σWt + νt) if and only if we take ν to be µ − 1 σ 2 . in that
dXt = σt dWt + µt dt. Moreover. Soluble SDEs are scarce. Such a solution to an SDE is called a diffusion. So o we have found one solution. In the stochastic world. and as it turns out. that d(ft gt ) = ft dgt + gt dft . that 2
1 Xt = exp σWt + µ − 2 σ 2 t . o 1 2 if we choose µ to be − 2 σ . o Suppose we are asked to solve the SDE
dXt = σXt dWt . which is what we wanted.
1 2
By applying Itˆ ’s formula to o
(Xt + Yt )2 − Xt2 − Yt2 = Xt Yt . for µt a general bounded integrable function of time?
The product rule
Another Newtonian law was the product rule. the only solution (up to constant multiples). So that is our guess. o
xxxx. 2
What does Itˆ tell us? That dXt is indeed σXt dWt . TNxxx
Exercise 3.
And again Itˆ conﬁrms our intuition. dYt = ρt dWt + νt dt. and this one is special enough to have a name: the Doleans exponential of Brownian motion. o
. In the more signiﬁcant case. Xt and Yt are adapted to the same Brownian motion W . there are two (seemingly) separate cases.
The ﬁnal term above is actually dXt dYt (following from (dWt )2 = dt) and again marks the difference between Newtonian and Itˆ calculus.
where σt and ρt are the respective volatilities of X and Y . Consider a simple twostep random walk:
. such as
dXt = σt dWt + µt dt. And thus by extension through their differentials. as will be explained in section 6.
Change of measure — the RadonNikodym derivative
To get some intuitive feel for the effects of a change of measure. Yet we don’t seem to mention measures in our stochastic differentials.48
CHAPTER 3. we should go back for a while to discrete processes. so do stochastic processes. As it happens. TNxxx
Exercise 3. µt and νt are their drifts. Wt ). not a manipulation of measure. But the only tool we have seen so far gives us no clue how Wt let alone Xt changes as the measure changes. At a deeper level these two stochastic cases can be reconciled by viewing Xt and ˜ Yt as both adapted to the twodimensional Brownian motion (Wt . ˜ and W and W are two independent Brownian motions. Brownian motions change in easy and pleasant ways under changes in measure. but a Brownian motion with respect to some measure P.6 Show that if Bt is a zerovolatility process and Xt is any stochastic process.
just as in the Newtonian case. but they are a manipulation of differentials of Brownian motion. And thus our stochastic differential formulation describes the behavior of the process X with respect to the measure P that makes the Wt (or of course the dWt ) a Brownian motion.
3.3. then
d(Bt Xt ) = Bt dXt + Xt dBt . ˜ dYt = ρt dWt + νt dt. CONTINUOUS PROCESSES
In the other case. Xt and Yt are two stochastic processes adapted to two different and independent Brownian motions. One of the central themes of the previous chapter was the importance of separating process and measure. We haven’t actually ignored the importance of measure — Wt is not strictly a Brownian motion per se. Here
d(Xt Yt ) = Xt dYt + Yt dXt . a PBrownian motion.4 Change of measure — the CMG theorem
Something remains hidden from us.
xxxx. We may have our basic tools for manipulating stochastic processes.
1. 0}. π2 . −2}. then (as long as all of them are strictly between 0 and 1) we know p1 . {0. Suppose we speciﬁed the probability of taking these paths: We could view this mapping of paths to path probabilities as encoding the measure P. 1. 0}. we can label each of the paths at the end with the π information encoding the measure. 1.9: Twostep recombinant tree Table 3.1: Path probabilities Path {0. −1. If we knew π1 . Thus if we represent our process with a nonrecombining tree. {0. π2 . π3 and π4 uniquely decides Q. we can follow four possible paths {0. −1. Again we can code this up with path probabilities.4. −2} Probability p 1 p2 p1 (1 − p2 ) (1 − p1 )p3 (1 − p1 )(1 − p3 ) =: π1 =: π2 =: π3 =: π4
To get from time 0 to time 2. 1. {0. π1 . −1. 0} {0. π2 . 2} {0. 2}.
.10: Tree with path probabilities marked
Now suppose we had a different measure Q with probabilities q1 . π3 and π4 . π3 and π4 .3. p2 and p3 . And again if each π is strictly between 0 and 1. say π1 . 0} {0. q2 and q3 . CHANGE OF MEASURE — THE CMG THEOREM
49
Figure 3. −1.
Figure 3.
so we have π1 . . then we have π1 . and thus dQ can’t exist. This is important enough to formalize. We dP could suppress those paths which had path probability zero. . π4 . Suppose one of the p’s is zero.50
CHAPTER 3. And vice versa. Not all the ratios πi /πi will be well deﬁned. And of course vice versa. . Somehow we can’t deﬁne dQ if Q allows something dP which P doesn’t. but none of the q ’s are. Formally.
In other words.11: Tree with RadonNikodym derivative marked
From dQ we can derive Q from P. there is a very natural way of encoding the differences between P and Q.
P(A) > 0 ⇐⇒ Q(A) > 0. CONTINUOUS PROCESSES
And with this encoding. . The second problem has a similar ﬂavor but is more serious. If we restrict ourselves to only providing πi for possible paths. so in some sense p2 really isn’t relevant. if A is possible under P then it is possible under Q. Equivalence Two measures P and Q are equivalent if they operate on the same sample space and agree on what is possible. dP What about pi or qi being zero or one? Two things happen — ﬁrstly it can become impossible to back out the pi from the πi . Those paths may have been Pimpossible but they are Qpossible.
Figure 3. then we have lost information about Q just where it is relevant — paths which are Qpossible. If we form the ratio πi /πi for each path i. . π2 . then we can recover the corresponding p’s. we write the mapping of paths to this ratio as dQ . But then of course. the paths corresponding to π1 and π2 are both impossible (probability zero). π4 . Then at least one πi will be zero when none of the πi are. Consider if p1 is zero then both π1 and π2 are zero and so information about p2 is lost. giving some idea of how to distort P so as to produce Q. If we throw them away. dP and dQ gives us the ratios πi /πi . π2 .
. but now we have lost something. . if A is any event in the sample space. How? If we have P. . and if A is impossible under P then it is also impossible under Q. . And thus Q. dP This random variable (random because it depends on the path) is called the RadonNikodym derivative of Q with respect to P up to time 2.
We speciﬁed X at this time and we only wanted an unconditioned expectation. dP
where T is the time horizon for dQ and XT is known at time T . but we would like a dP process. In formal terms. Thus two measures P and Q must be equivalent before they will have RadondP Nikodym derivatives dQ and dQ . One of the reasons for deﬁning it was the efﬁcient coding it represented. CHANGE OF MEASURE — THE CMG THEOREM
51
dP We can only meaningfully deﬁne dQ and dQ if P and Q are equivalent. dP
Just like X . dP
Expectation and
dQ dP
While we are still working with discrete processes. So the expectation of X with respect to P is given by
EP (X) =
i
πi xi . Everything we needed to know about Q could be extracted from P and dQ . we should stock up on some facts about expectation and the RadonNikodym derivative. The claim X is a random variable. and then dP only where paths are Ppossible. ζt is the RadonNikodym derivative dQ but only following paths up to time t. or in other words a mapping from paths to values — we can let xi denote the value the claim takes if path i is followed. in this case T = 2.4. it represents just one simple case: dQ is deﬁned with a dP particular time horizon in mind — the ends of the paths. the result we derived was
EQ (XT F0 ) = EP dQ XT F0 . And the conversion from Q to P is pleasingly simple: EQ (X) = EP dQ X . is a random variable which we can take the expectation of. That is.
where i ranges over all four possible paths. But of course if paths are Pimpossible then we know how Q acts on those paths — if Q is equivalent to P then they are Qimpossible as well. and setting ζt to be the RadonNikodym derivative taken up to the horizon t.3. dP Attractive though this is. dP Consider then a claim X known by time 2 on our discrete twostep process. And the expectation of X with respect to Q is
EQ (X) =
i dQ dP
πi xi =
i
π
πi xi πi
= EP
dQ X .
RadonNikodym process
We can do this by letting the time horizon vary. and only looking at the ratio of dP
. What about EQ (Xt Fs ) dP for t not equal to T and s not equal to zero? We need somehow to know dQ not just dP for the ends of paths but everywhere — dQ is a random variable.
Concretely.
Figure 3. where Xt is a claim known at time t.8 Prove this on the tree.7 Prove that this equation holds for t = 0.
. The process ζt represents just what we wanted — an idea of the amount of change of measure so far up to time t along the current path. dP
for every t less than or equal to the horizon T . we can ﬁll in ζt on our tree in terms of the p’s and q ’s (ﬁgure 3.
xxxx.
ζt = EP dQ Ft . 1. CONTINUOUS PROCESSES
probabilities up to that time.12: Tree with ζt process marked (¯i = 1 − pi . That is. TNxxx
Exercise 3.
xxxx. We can see that the expectation with respect to P unpicks the dQ in just the right dP way. In other words
−1 EQ (Xt Fs ) = ζs EP (ζt Xt Fs ). the derivative process is just 1. qi = 1 − qi ) p ¯
In fact there is another expression for ζt as the conditional expectation of the T horizon RadonNikodym derivative. For instance.12).52
CHAPTER 3. If we wanted to know EQ (Xt ) it would be EP (ζt Xt ). 2. the possible paths are {0. the change up to time t with the change up to time s removed. −1} and the derivative ζ1 has values on them of q1 /p1 and (1 − q1 )/(1 − p1 ) respectively. TNxxx
Exercise 3. If we want to know EQ (Xt Fs ) then we need the amount of change of measure from time s to time t — which is just ζt /ζs . as the only ‘path’ is the point {0} which has probability 1 under both P and Q. At time zero. 1} and {0. at time 1.
such that dP (i) EQ (XT ) = EP
dQ XT dP . t1 . 1) random variable X . CHANGE OF MEASURE — THE CMG THEOREM
'
53 $
RadonNikodym summary Given P and Q equivalent measures and a time horizon T . tn−1 .4. . 1] is the integral of the density over the interval. One approach is to specify a path if not for all times before the horizon T . Standard probability theory gives some clue to the technology required. . 2π
In some loose sense. . can be represented via the density fP (x). But marginal distributions aren’t enough — a single marginal distribution won’t capture the nature of the process (we can see that clearly even on a discrete tree). . by means of some conceptual handle on a particular path speciﬁed for all times t < T . 01 fP (x)dx. fP (x) represents the relative likelihood of the event {X = x} occurring. 2π
For example. . we can deﬁne a random variable dQ deﬁned on Ppossible paths. Consider then. where
1 2 1 fP (x) = √ e− 2 x . taking positive real values. then at least for some arbitrarily large yet still ﬁnite set of times {t0 = 0. Despite the continuous nature of the state space.
Change of measure — the continuous RadonNikodym derivative
What now? To deﬁne a measure for Brownian motion it seems we have to be able to write down the likelihood of every possible path the process can take. We need to capture the idea of a likelihood of a path in the continuous case. tn = T }. ranging across not only a continuousvalued state space but also a continuousvalued time line.3413. For example.
for all claims XT knowable by time T . xn } at times {t1 . .
%
−1 (ii) EQ (Xt Fs ) = ζs EP (ζt Xt Fs ). if we were content merely to represent the marginal distributions for the process at each time. . . . . the probability that X takes a value in some subset A of the reals is
P(x ∈ A) =
A
1 2 1 √ e− 2 x dx. then we could write
. the set of paths which go through the points {x1 . If there were just one time t1 and one point x1 . which has value 0. In exact terms. Or. corresponding to a normal N (0. tn }. the chance of X being in the interval [0. .
s ≤ t ≤ T. We need nothing less than all the marginal distributions at each time t conditional on every history Fs for all times s < t.
&
where ζt is the process EP
dQ dP
Ft . Nor will all the marginal distributions for each time t. the measure P on the real numbers. .3. less informally the probability that X lies between x and x + dx is approximately fP (x)dx . we know that we can express likelihoods in terms of a probability density function.
then given the third condition of Brownian motion that increments ∆Wi = W (ti ) − W (ti−1 ) are mutually independent. Just as the measure P can be approached through a limiting time mesh. . . dP A→{ω} P(A)
. . n}. then we can for ﬁnitely many ti . .
So we can write down a likelihood function corresponding to the measure P for a process on a ﬁnite set of times. we have a handle on the measure P for a continuous process. i = 1. . which is the density function of a normal N (0. . . . The RadonNikodym derivative can be thought of as the limit
dQ Q(A) (ω) = lim . . Wtn ) is in A is exactly the integral over A of the n likelihood function fP . t3 }
Joint likelihood function for Brownian motion If we take t0 and x0 to be zero. . fP (x). tn }. . The event of paths agreeing with ω on the mesh. so can the RadonNikodym derivative dQ . . CONTINUOUS PROCESSES
down the likelihood of such a path. or
1 fP (x) = √
x2 1 exp − 2t1 2πt1
. xn } at times {t1 . .
And if we can do this for one time t1 . gets smaller and smaller till it is just the single pointset {ω}. . .54
CHAPTER 3. dP A = {ω : Wti (ω ) = Wti (ω). . We could use the probability density function of 1 Wt1 . . If A is some subset of Rn . t1 ).13: Two Brownian motions agreeing on the set {t1 . then the Pprobability that the random nvector (Wt . And in the continuous limit. xn ) for the process taking values {x1 . . . . we can write down
n n fP (x1 . xn )
=
i=1
√
(∆xi )2 1 exp − 2∆ti 2π∆ti
. and write ∆xi for xi − xi−1 and ∆ti = ti − ti−1 . .
Figure 3. . t2 . . All we n require is the joint likelihood function fP (x1 . .
for every ordered time mesh {t1 . Consider Wt a PBrownian motion. . Given a path ω . T ) with respect to P. which tells us that it is the same as the Pexpectation EP dQ exp(θWT ) . What does Wt look like with respect to Q? One place to start. . . This continuoustime derivative dQ still satisﬁes the results that dP (i) EQ (XT ) = EP
dQ XT . dP s ≤ t ≤ T.4. for example. is to look at the marginal of WT under Q. . then (out of nowhere) deﬁne Q to be a measure equivalent to P via
dQ = exp −γWT − 1 γ 2 T . dP This equals
1 1 EP exp(−γWT − 2 γ 2 T + θWT ) = exp − 2 γ 2 T + 1 (θ − γ)2 T . . . 2 dP
for some time horizon T .
where ζt is the process EP Ft . and it is just a start. we had a PBrownian motion Wt . . Suppose. We need to ﬁnd the likelihood function of WT with respect to Q. T ]. . . xn ) dP P
as the mesh becomes dense in the interval [0. What does Wt look like under an equivalent measure Q — is it still recognizably Brownian motion or something quite different? Foresight can provide one simple example. 2
because WT is a normal N (0. CHANGE OF MEASURE — THE CMG THEOREM
55
RadonNikodym derivative — continuous version Suppose P and Q are equivalent measures.
−1 (ii) EQ (Xt Fs ) = ζs EP (ζt Xt Fs ).
. we deﬁne xi to be Wti (ω).3. and Xt is any process adapted to the history
Simple changes of measure — Brownian motion plus constant drift
We have the mechanics of change of measure but still no clue about what change of measure does in the continuous world. tn } (with tn = T ).
dQ dP
Ft .
To calculate EQ (exp(θWT )). . . . or something equivalent. One useful trick is to look at momentgenerating functions: Identifying normals A random variable X is a normal N (µ. xn ) dQ (ω) = lim n . σ 2 ) under a measure P if and only if
EP (exp(θX)) = exp θµ + 1 θ2 σ 2 . n→∞ f (x1 . 2
for all real θ. we can use fact (i) of the RadonNikodym derivative summary. and then the derivative dQ dP up to time T is deﬁned to be the limit of the likelihood ratios
n fQ (x1 .
a lot of other process also have a marginal normal N (−γT. such a path is much more likely. under Q. and the chances are that is what we see. it is Ppossible. t) under Q. dP
˜ That both Wt and Wt are Brownian motion. What about Wt for t less than T ? The marginal distribution of WT is what we would expect if Wt under Q were a Brownian motion plus a constant drift −γ . And so it is. albeit with respect to different measures. Of course. Correspondingly. W might follow a path which drifts downwards for a time at a rate of about −γ . TNxxx
Fs = exp
. seems paradoxical. But switching from P to Q just changes the relative likelihood of a particular path being chosen. This is just the consequence of the common sense thought that paths which end up negative are more likely under Q (Brownian motion plus downward drift) than they are under P (driftless Brownian motion). We can see this in the RadonNikodym derivative dQ which is large when WT dP is very negative.
. The process Wt is a Brownian motion with respect to P and Brownian motion with constant drift −γ under Q.56
CHAPTER 3.
1 2 2θ t
˜ ˜ (iii) EQ exp θ (Wt+s − Wt )
xxxx. on the other hand. and small when WT is closer to zero or positive.
Exercise 3. Although that path is Punlikely. but it would be an elegant result if the sole effect of changing from P to Q via dQ 1 2 dP = exp −γWT − 2 γ T were just to punch in a drift of −γ .
which is the momentgenerating function of a normal N (−γT. But it still could be just improbable Brownian motion behavior. we can prove dP ˜ the three conditions for Wt = Wt + γt to be QBrownian motion:
˜ ˜ (i) Wt is continuous and W0 = 0. ˜ ˜ (iii) Wt+s − Wt is a normal N (0. Thus the marginal distribution of WT . For example. ˜ (ii) Wt is a normal N (0.9 Show that (ii) and (iii) are equivalent to (ii) and (iii) respectively. CONTINUOUS PROCESSES
Simplifying the algebra. is also a normal with variance T but with mean −γT . Using our two results about dQ . and prove them using the change of measure process
ζt = EP dQ Ft . t) independent of Fs
The ﬁrst of these is true and (ii) and (iii) can be reexpressed as
˜ (ii) EQ exp(θWt ) = exp
1 2 2θ t
. we have
1 EQ (exp(θWT )) = exp −θγT + 2 θ2 T . T ) distribution at time T . T ). Under Q. paths which ﬁnish near or above zero are less likely under Q than P.
Within limits. Wt plus drift γt is QBrownian motion. Additionally the RadonNikodym derivative of Q with respect to P (at time T ) is T 1 T 2 exp − 0 γt dWt − 2 0 γt dt . This is what our theorem provides. drift is one of the elements of our stochastic differential form of processes.3. Now we can see the rewards of our Brownian calculus instantly — CMG becomes a powerful tool for controlling the drift of any process. Within constraints. if we want to turn a PBrownian motion Wt into a Brownian motion with some speciﬁed drift −γt .
. drift is measure and measure drift.
In other words. CHANGE OF MEASURE — THE CMG THEOREM
57
CameronMartinGirsanov
So this one change of measure just changed a vanilla Brownian motion into one with drift — nothing else. but all our processes are disguised Brownian motions at heart. And of course. then there exists an F previsible process γt such that
˜ Wt = Wt +
t 0
γs ds
is a QBrownian motion.4. That is. CameronMartinGirsanov converse If Wt is a PBrownian motion. then there exists a measure Q such that (i) Q is equivalent to P (ii)
dQ = exp − dP
T 0
γt dWt −
1 2
T 0
2 γt dt
˜ (iii) Wt = Wt +
t 0 γs ds
is a QBrownian motion. then there’s a Q which does it. CameronMartinGirsanov theorem If Wt is a PBrownian motion and γt is an F previsible process satisfying the 1 T 2 boundedness condition EP exp 2 0 γt dt < ∞. In fact all that measure changes on Brownian motion can do is to change the drift. and Q is a measure equivalent to P. All the processes that we are interested in are representable as instantaneous differentials made up of some amount of Brownian motion and some amount of drift. The mapping of stochastic differentials under P to stochastic differentials under Q is both natural and pleasing. Conversely to the theorem.
CMG and stochastic differentials
The CMG theorem applies to Brownian motion. Wt is a drifting QBrownian motion with drift −γt at time t.
58
CHAPTER 3. CONTINUOUS PROCESSES
Suppose that X is a stochastic process with increment
dXt = σt dWt + µt dt,
where W is a PBrownian motion. Suppose we want to ﬁnd if there is a measure Q such that the drift of process X under Q is νt dt instead of µt dt. As a ﬁrst step, dX can be rewritten as
dXt = σt dWt + µt − νt σt dt + νt dt.
If we set γt to be (µt − νt )/σt , and if γ then satisﬁes the CMG growth condition 1 T 2 ˜ EP exp 2 0 γt dt < ∞ then indeed there is a new measure Q such that Wt :=
Wt + − νs )/σs ds is a QBrownian motion. But this means that the differential of X under Q is ˜ dXt = σt dWt + νt dt, ˜ where W is a QBrownian motion — which gives X the drift νt we wanted. We can also set limits on the changes that changing to an equivalent measure can wreak on a process. Since the change of measure can only change the Brownian motion to a Brownian motion plus drift, the volatility of the process must remain the same.
t 0 (µs
Examples — changes of measure
1. Let Xt be the drifting Brownian process σWt + µt, where W is a PBrownian motion and σ and µ are both constant. Then using CMG with γt = µ/σ , there ˜ ˜ exists an equivalent measure Q under which Wt = Wt + (µ/σ)t and W is a Q˜ Brownian motion up to time T . Then Xt = σ Wt , which is (scaled) QBrownian motion. The measures also give rise to different expectations. For example, EP (Xt2 ) equals µ2 t2 + σ 2 t, but EQ (Xt2 ) = σ 2 t. 2. Let Xt be the exponential Brownian motion with SDE
dXt = Xt (σdWt + µdt),
where W is PBrownian motion. Can we change measure so that X has the new
SDE
dXt = Xt (σdWt + νdt),
for some arbitrary constant drift ν ? Using CMG with γt = (µ − ν)/σ , there is indeed a measure Q under which ˜ Wt = Wt + (µ − ν)t/σ is a QBrownian motion. Then X does have the SDE
˜ dXt = Xt (σdWt + νdt), ˜ where W is a QBrownian motion.
3.5. MARTINGALE REPRESENTATION THEOREM
59
3.5 Martingale representation theorem
We can solve some SDEs with Itˆ ; we can see how SDEs change as measure changes. o But central to answering our pricing question in chapter two was the concept of a measure with respect to which the process was expected to stay the same, the martingale measure for our discrete trees. The price of derivatives turned out to be an expectation under this measure, and the construction of this expectation even showed us the trading strategy required to justify this price. And so it is here. First the description again: Martingales A stochastic process Mt is a martingale with respect to a measure P if and only if (i) EP (Mt ) < ∞, (ii) EP (Mt Fs ) = Ms , for all t for all s ≤ t.
The ﬁrst condition is merely a technical sweetener, it is the second that carries the weight. A martingale measure is one which makes the expected future value conditional on its present value and past history merely its present value. It isn’t expected to drift upwards or downwards. Some examples: (1) Trivially, the constant process St = c (for all t) is a martingale with respect to any measure: EP (St Fs ) = c = Ss , for all s ≤ t, and for any measure P. (2) Less trivially, PBrownian motion is a Pmartingale. Intuitively this makes sense — Brownian motion doesn’t move consistently up or down, it’s as likely to do either. But we should get into the habit of checking this formally: we need EP (Wt Fs ) = Ws . Of course we have that the increment Wt − Ws , is independent of Fs and distributed as a normal N (0, t − s), so that EP (Wt − Ws Fs ) = 0. This yields the result, as
EP (Wt Fs ) = EP (Ws Fs ) + EP (Wt − Ws Fs ) = Ws + 0
(3) For any claim X depending only on events up to time T , the process Nt = EP (XFt ) is a Pmartingale (assuming only the technical constraint EP (X) < ∞). Example (3) is an elegant little trick for producing martingales — and as we shall see (and have already seen in chapter two) central to pricing derivatives. First why? Convince yourself that Nt = EP (XFt ) is a welldeﬁned process — the ﬁrst stage of the alchemy is the introduction of a time line into the random variable X . Now for Nt to be a Pmartingale, we require EP (Nt Fs ) = Ns , but for this we merely need to be satisﬁed that
EP EP (XFt )Fs = EP (XFs ).
60
CHAPTER 3. CONTINUOUS PROCESSES
That is, that conditioning ﬁrstly on information up to time t and then on information up to time s is just the same as conditioning up to time s to begin with. This property of conditional expectation is the tower law.
xxxx; TNxxx
Exercise 3.10 Show that the process Xt = Wt +γt , where Wt is a PBrownian motion, is a Pmartingale if and only if γ = 0.
Representation
In chapter two, we had a binomial representation theorem — if Mt and Nt are both Pmartingales then they share more than just the name — locally they can only differ by a scaling, by the size of the opening of each particular branching. We could represent changes in Nt by scaled changes in the other nontrivial Pmartingale. Thus Nt itself can be represented by the scaled sum of these changes. In the continuous world: Martingale representation theorem Suppose that Mt is a Qmartingale process, whose volatility σt satisﬁes the additional condition that it is (with probability one) always nonzero. Then if Nt is any other Qmartingale, there exists an F previsible process φ such that T 2 2 0 φt σt dt < ∞ with probability one, and N can be written as
t
Nt = N0 +
0
φs dMs .
Further φ is (essentially) unique. This is virtually identical to the earlier result, with summation replaced by an integral. As we are getting used to, the move to a continuous process extracts a formal technical penalty. In this case, the Qmartingale’s volatility must be positive with probability 1 — but otherwise our chapter two result has carried across unchanged. If there is a measure Q under which Mt is a Qmartingale, then any other Qmartingale can be represented in terms of Mt . The process φt is simply the ratio of their respective volatilities.
Driftlessness
We need just one more tool. Thrown into the discussion of martingales was the intuitive description of a martingale as neither drifting up or drifting down. We have, though, a technical deﬁnition of drift via our stochastic differential formulation, An obvious question springs to mind: are stochastic processes with no drift term always martingales, and vice versa can martingales always be represented as just σt dWt for some F previsible volatility process σt ? Nearly.
a driftless process may not be a martingale. show that dXt = σt Xt dWt is a Pmartingale. then
X is a martingale ⇐⇒ X is driftless (µt ≡ 0). E practical) test is: A collector’s guide to exponential martingales If dXt = σt Xt dWt .
xxxx.11 If σt is a bounded function of both time and sample path. The condition (in this case.6. but for these speciﬁc exponential examples. If a process Xt is a Pmartingale then with Wt a PBrownian motion. The other way round is true (up to a technical constraint).6 Construction strategies
We have the mathematical tools — Itˆ . Such processes are called local martingales.
If the technical condition fails. CameronMartinGirsanov. and the martino gale representation theorem — now we need some idea of how to hook them into
. but harder. a better (more
< ∞ =⇒ X is a martingale.
This is just the integral form of the increment dXt = φt dWt .
3.3. we have an F previsible process φt such that
t
Xt = X0 +
0
φs dWs .
Exponential martingales
The technical constraint can be tiresome. which has no drift term. For reference: A collector’s guide to martingales If X is a stochastic process with volatility σt (that is dXt = σt dWt + µt dt) which satisﬁes the technical condition E
T 0 2 σs ds
1 2
< ∞. take the (driftless) SDE for an exponential process dXt = σt Xt dWt . For example. for some F previsible process σt . TNxxx
Exercise 3.
t 0 σs dWs
We also note that the solution to the SDE is Xt = X0 exp
−
1 t 2 2 0 σs ds
. CONSTRUCTION STRATEGIES
61
One way round we can do for ourselves with the martingale representation theorem. then
E exp 1 2
T 0 2 σs ds T 0 2 2 σs Xs ds
1 2
<
∞) is difﬁcult to check.
and in the continuous case it is equivalent to an SDE. The security component of the portfolio φ should be F previsible: depending only on information up to time t but not t itself. If φ were leftcontinuous (that is.
. ψt ) is a portfolio with stock price St and bond price Bt . Vt . ψt ) is selfﬁnancing ⇐⇒ dVt = φt dSt + ψt dBt . There is an intuitive way to think about previsibility.
What SDE?
With stock price St and bond price Bt . we’ll have a market consisting of one random security and a riskless cash account bond. A portfolio is selfﬁnancing if and only if the change in its value only depends on the change of the asset prices. At the next time instant. and the old portfolio has to be adjusted to give a new portfolio as instructed by the trading strategy (φ. BlackScholes for example. If φ were only rightcontinuous (that is φs tends to φt only as s tends downwards to t from above). ψ) A portfolio is a pair of processes φt and ψt which describe respectively the number of units of security and of the bond which we hold at time t. If the cost of the adjustment is perfectly matched by the proﬁts or losses made by the portfolio then no extra money is required from outside — the portfolio is selfﬁnancing. In our discrete language.
Selfﬁnancing strategies
With the idea of a portfolio comes the idea of a strategy. two things happen: the old portfolio changes value because St and Bt have changed price. of a portfolio (φt . ψ). ψt ) at time t is given by Vt = φt St + ψt Bt . The description (φt . In the simplest models. The processes can take positive or negative values (we’ll allow unlimited shortselling of the stock or bond). then φ need not be. And one particularly interesting set of strategies or portfolios are those that are ﬁnancially selfcontained or selfﬁnancing. and with this comes the idea of a portfolio.
In continuous time. In the discrete framework this was captured via a difference equation. CONTINUOUS PROCESSES
a ﬁnancial model. ψt ) is a dynamic strategy detailing the amount of each component to be held at each instant. φs tends to φt as s tends upwards to t from below) then φ would be previsible.62
CHAPTER 3. The portfolio (φ. the value. we get a stochastic differential equation: Selfﬁnancing property If (φt . then
(φt . we had the difference equation
∆Vi = φi ∆Si + ψi ∆Bi .
TNxxx
Exercise 3. but it will all be due to ﬂuctuation of the stock. as of now. dVt = 2Wt dWt which o is identical to φt dSt + ψt dBt as required. and the value Vt = φt St + ψt Bt = Wt2 − t. By Itˆ ’s formula. The second example should convince us that being selfﬁnancing is not an automatic property of a portfolio.
xxxx. φt is previsible. Checking this formally. What kind of portfolios are selfﬁnancing? (1) Suppose φt = ψt = 1 for all t.12 Verify the Itˆ claim in (2) above (which also shows that Wt2 −t o is a martingale). A replicating strategy for X is a selfﬁnancing portfolio (φ. Why should we care about replicating strategies? For the same reason as we wanted them in the discrete market models. ψ) such that T 2 2 0 σt φt dt < ∞ and VT = φT ST + ψT BT = X . ψt ) portfolio ought to be selfﬁnancing. The (φt . Vt = Wt + 1 implies that dVt = dWt which is the same as φt dSt + ψt dBt . Surprising though it seems: holding as many units of stock as twice its current price. (2) Suppose φt = 2Wt and ψt = −t − Wt2 . The claim X gives the value of some derivative which we need to pay off at time T . as we required (remembering that dBt = 0).6. CONSTRUCTION STRATEGIES
63
Suppose the stock price St is given by a simple Brownian motion Wt (so St = Wt for all t). and the bond price Bt is constant (Bt = 1 for all t). Every time we claim a portfolio is selfﬁnancing we have to turn the handle on Itˆ ’s formula to check the SDE. Here (φt . ψt ) is a portfolio. ψt ) strategy could (in a perfect market) be followed to our heart’s content without further funding. given a model for S and B . though a rollercoaster strategy. no extra money is needed to come in to uphold the (φt . Intuitively. is exactly offset by the stock proﬁts and the changing bond holding of −(t + Wt2 ). ψt ) strategy would have required injections or forced outﬂows of cash. ψt ) strategy and none comes out — this (φt . The Itˆ check worked. then the value of the portfolio (Vt = Wt + 1) may ﬂuctuate.
.3. If we hold a unit of stock and a unit of bond for all time without change. but it could easily have failed if o ψt had been different — the (φt . We want a price if there is one. o
Trading strategies
Now we can deﬁne a replicating strategy for a claim: Replicating strategy Suppose we are in a market of a riskless bond B and a risky security S with volatility σt and a claim X on events up to time T .
ψt ) . We can lay out a battle plan. And of course if the derivative price had been higher than Vt . We need a model for the behavior of the stock — simple enough that we actually can ﬁnd replicating strategies but not so simple that we can’t bring ourselves to believe in it as a model of the real world. the bought derivative and sold portfolio would safely cancel at time T . St = S0 exp(σWt + µt). our market will consist of a riskless
. we can price derivatives in the model. ψ) is selfﬁnancing and the portfolio is worth X at time T guaranteed. And if we can. no risk means no fear. (And speciﬁcally. and then use the martingale representation theorem (section 3. as usual with arbitrage.5) to create a replicating strategy for each claim. continuing to be short (φ. using whatever tools we have to hand we ﬁnd replicating strategies for all useful claims X . and no extra money is required between times t and T . µ and σ such that the bond price Bt and the stock price follow
Bt = exp(rt).64
CHAPTER 3. tie down the price of the claim X not just at payoff but everywhere. We have the tools and we’ve seen the overall approach at the end of chapter two.7 BlackScholes model
We need a model to cut our teeth on. Itˆ will oil the works. we will use the CameronMartinGirsanov theorem (section 3.
where r is the riskless interest rate. σ is the stock volatility and µ is the stock drift. So taking the stock model of section 3. ψ) to the same effect. one unit could have been many. Then. The rest of the book consists of upping the stakes in complexity of models and of claims.4) to change it into a martingale. The proﬁt created by the mismatch at time t can be banked there and then without risk. CONTINUOUS PROCESSES
If there is a replicating strategy (φt . ψ) until time T . o
The model
Our ﬁrst model — basic BlackScholes We will posit the existence of a deterministic r. then we could have sold the derivative and bought the selfﬁnancing (φ. There are no transaction costs and both instruments are freely and instantaneously tradable either long or short at the price quoted. if they exist. Following in Black and Scholes’ footsteps.
3. Because (φ. Replicating strategies. the price at time zero is V0 = φ0 S0 + ψ0 B0 ) If it were lower. We deﬁne a market model with a stock price process complex enough to satisfy our need for realism.1. And. then the price of X at time t must be Vt = φt St + ψt Bt . a market player could buy one unit of the derivative at time t and sell φt units of S and ψt units of B against it.
The problems it causes are more tedious than fatal — as we’ll see soon. Yt = log(St ). St = exp(σWt + µt ). the ﬁrst thing to do is to lull the drift in this SDE . But we’ll temporarily simplify things. 2
In order for St to be a martingale. (2) Form the process Et = EQ (XFt ). The stock follows an exponential Brownian motion.
& %
The tools described earlier on will be essential to do this.
' $
Three steps to replication (1) Find a measure Q under which St is a martingale. We shall use the CameronMartinGirsanov theorem (section 3. it’s the interest rate r. we want to see if we can ﬁnd a replicating strategy (φt .
Zero interest rates
If there’s one parameter that throws up a smokescreen around a ﬁrst run at an analysis of the BlackScholes model.
Step one
For two different reasons — ﬁrstly we need to apply the CameronMartinGirsanov theorem. For an arbitrary claim X . so the logarithm of the stock price. BLACKSCHOLES MODEL
65
constantinterest rate cash bond and a risky tradable stock following an exponential Brownian motion. and set r to be zero.4) for the ﬁrst step and the martingale representation theorem (section 3. Itˆ makes it possible to write down the SDE for St = exp(Yt ) as o
dSt = σSt dWt + µ + 1 σ 2 St dt. of course. If we let γt be a process with constant value γ = µ + 1 σ 2 /σ . And as we shall see here. ψt ). (3) Find a previsible process φt . then the CM2 ˜ t = Wt + γt is QBrownian G theorem says that there is a measure Q such that W
.7. the tools we have are powerful enough to cope. it is at least a plausible match to the real world. it is quite hard enough to start with. knowable by some horizon time T . So now we begin. But.
Finding a replicating strategy
We shall follow a threestep process outlined in this box here. As we’ve seen in section 3. such that dEt = φt dSt .3.1. follows a simple drifting Brownian motion Yt = σWt + µt . Thus the SDE for Yt is easy to write down: dYt = σdWt + µdt .5) for the third one. and secondly we need to be able to tell if St is a Qmartingale for a given Q — we want to ﬁnd an SDE for St .
from the martingale representation theorem. dEt = φt dSt . we can invoke the martingale representation theorem. but this is true because the volatility is just σSt . under which both Et and St are Qmartingales.
Replicating strategy
Our strategy is to:
• hold φt units of stock at time t and • hold ψt = Et − φt St units of the bond at time t. ψt ). (The technical boundedness condition is satisﬁed because γt is constant. Thus dVt = dEt . (To use the theorem.5. Consequently.
No drift term. dEt = φt dSt for some φt . So the martingale representation theorem tells us an important fact: given a Q that makes St a Qmartingale with positive volatility.
. The exponential martingales box (section 3. Q is the martingale measure for St . setting ψt to be the only thing it can be. we can convert X into a process by forming Et = EQ (XFt ).
Is it selfﬁnancing? The value of the portfolio at time t is
Vt = φt St + ψt Bt = Et .
Step two
Given Q. given that we want the portfolio to be worth Et for all t.
or.) Substituting in. CONTINUOUS PROCESSES
motion. but of course dEt = φt dSt . This is.5) contains a condition in terms of σ for St to be a martingale under Q. the SDE is now
˜ dSt = σSt dWt . a Qmartingale. As σ is constant. So we should try it. and it’s tempting to believe that we have got one half of it.
because the bond Bt is constantly equal to 1. Since dBt is zero. and both σ and St are always positive. as we have already discussed in example (3) of section 3.66
CHAPTER 3. thus St could be a Qmartingale. There exists a previsible process φt which constructs Et = EQ (XFt ) out of St .) Formally:
t
Et = EQ (XFt ) = EQ (X) +
0
φs dSs . we need to check that the volatility of St is always positive. We need a replicating strategy (φt . of course. namely dVt = φt dSt + ψt dBt . we have the selfﬁnancing condition we want. the condition holds which means that St must be a Qmartingale.
Step three
Since there is a Q.
and considered a forward contract with claim ST − k for some price k . So we might expect to fail in our search for replicating strategies. BLACKSCHOLES MODEL
67
Since the terminal value of the strategy VT is ET = X . It is after all particularly odd that despite the lack of knowledge about the claim’s eventual value. What happens if r is nonzero? We can’t just ignore it. Since the claim depends on St . If we actually want to crank the handle and calculate derivative prices for a particular claim. 2 The third surprise is the simplicity of the process St under its martingale measure. of some function of the values of St up to t = T . but not the expectation in a traditional statistics sense. But St is also an exponential Brownian motion under Q. The second surprise. So if we felt that St was tractable under its original measure. The ﬁrst is just the fact that there are replicating strategies for arbitrary claims. we have a replicating strategy for X — which means there is an arbitrage price for X at all times. this normally involves calculating the expected value. we have to be able to calculate the expected value of the claim under the martingale measure Q. If we solve the SDE. which is the measure that makes it an exponential Brownian motion with drift µ and volatility σ . is that the price of the derivative has such a simple expression — the expected value of the claim. under Q. We already know that the k which gives the forward contract
. then this task could be unpleasant too. then
1 ˜ St = exp σ Wt − 2 σ 2 t . And arbitrage always wins out. we can nevertheless trade in the market in such a way that we always produce it. the price of the claim X is its expected value under the measure that makes the stock process St a martingale. Speciﬁcally there is an arbitrage price for X at time zero — the value of the (φt . it doesn’t give a price. The price happens to be an expectation. It is worth pausing to let a few surprises sink in. There is a replicating strategy and thus an arbitrage price for the claim. ψt ) portfolio at time zero. it is also tractable under the martingale measure. All that expectation could give us would be a longterm average of the claim’s payout.3. It is the easiest thing to forget that this is not the expectation of the claim with respect to the real measure of St . which makes the price E0 . It could only be the expectation if quite by chance the drift µ we believe in for the stock were exactly and precisely right to make St a martingale in the ﬁrst place (µ = − 1 σ 2 ).
and we ﬁnd that St has the same constant volatility σ and a new but also constant 1 drift of − 2 σ 2 . And though that could be a useful thing to know in order to judge whether punting with the derivative is worthwhile in the long run. The model that we have chosen isn’t too unrealistic — it has the right kind of behavior and a healthy degree of randomness. If St were an unpleasant process under Q. Suppose we did.
Nonzero interest rates
Now we can bring the interest rate r back in again. and just as important.7. In other words. or EQ (X).
and form a discounted stock −1 −1 Zt = Bt St and a discounted claim BT X .
So Zt . this is all just heuristic justiﬁcation.
Step two
We need a process which hits the discounted claim and is also a Qmartingale. we can. So there 2 ˜ exists (another) Q equivalent to the original measure P and a QBrownian motion Wt such that
˜ dZt = σZt dWt . If we can’t ﬁnd a replicating strategy then. attractive as our guess is. as before. So maybe our analysis will work again. The arbitrage to produce this is easy to ﬁgure out.
Step three
The discounted stock price Zt is a Qmartingale.
Even discounting the claim won’t help in this case. We call Bt the discount process. In fact. of simply taking the expected value of the claim under the martingale measure for St cannot work. conditional expectation provides it. Thus the martingale representation theorem gives us a previsible φt such that dEt = φt dZt . namely by forming the process Et = −1 EQ (BT XFt ).
EQ ST − S0 erT = S0 1 − erT = 0.
.
xxxx. and the proof is only in the doing. And. In this discounted world. If the growth of cash is annoying. So our rule of ﬁnding a measure which makes St into a martingale only holds true when r is zero. simply remove it by dis−1 counting everything. When r is not zero. it is not too hard to write down an SDE
1 dZt = Zt σdWt + (µ − r + 2 σ 2 )dt . CONTINUOUS PROCESSES
a zero value at time zero is k = S0 erT . TNxxx
Exercise 3.68
CHAPTER 3. the inexorable growth of cash gets in the way. But our rule. So we take a guess.
Step one
To make Zt into a martingale. we could be forgiven for thinking that r was zero again. Focusing on our discounted stock process Zt . Fortunately. it is also wrong. we can invoke CMG just as before. and so is the conditional expectation process of the discounted claim Et . when r was zero. is driftless and a martingale. Of course. under Q.13 Prove it. only now to introduce a drift of (µ − r + 1 σ 2 )/σ to the underlying Brownian motion.
Selfﬁnancing strategies A portfolio strategy (φt . we can write dVt as
dVt = Bt dEt + Et dBt . So our replicating strategy is to
• hold φt units of the stock at time t. so we can try that in the real world too. then we have proved that (φt .6 again.7. we have
dVt = φt dSt + ψt dBt .
. and so dVt = φt Bt dZt + Et dBt . ψt ) of holdings in a stock St and a nonvolatile cash bond Bt has value Vt = φt St + ψt Bt and discounted value Et = φt Zt + ψt . Thus following exercise 3.
But dEt is φt dZt (our fact from the martingale representation theorem).
or equivalently
dEt = φt dZt . A bit of rearrangement tells us that Et = φt Zt + ψt . (φt . What about the bond holding? The bond holding in the discounted world is ψt = Et − φt Zt . and • hold ψt = Et − φt Zt units of the bond.
But. Some reassurance comes from the fact that at time T we will be holding φT units of the stock and ψT units of the bond which will be worth φT ST + ψT BT = BT ET = X . let us try φt out in the real world as well. Since we know that VT = X .3.
Are we right? The value Vt of the portfolio (φt . Our guesses came good. from exercise 3.6. ψt ) is selfﬁnancing. Then the strategy is selfﬁnancing if either
dVt = φt dSt + ψt dBt . ψt ) is given by Vt = φt St + ψt Bt = Bt Et . So just as a guess. and thus
dVt = φt Bt dZt + (φt Zt + ψt )dBt = φt (Bt dZt + Zt dBt ) + ψt dBt . ψt ) is a replicating strategy for X .
That is. or equivalently if changes in its discounted value are due only to changes in the discounted values of the assets. d(Bt Zt ) = Bt dZt + Zt dBt . but in our shadow discounted world we can hit the discounted claim by holding φt units of the discounted stock. and since St = Bt Zt .
A strategy is selfﬁnancing if changes in its value are due only to changes in the assets’ values. where Z is −1 the discounted stock process Zt = Bt − St . BLACKSCHOLES MODEL
69
We want to hit the real claim with amounts of the real stock.
%
The important measure Q is not the measure which makes the stock a martingale. Following Black and Scholes. µ and σ such that their respective prices can be represented as St = S0 exp(σW t + µt) and Bt = exp(rt). But how do we ﬁnd this? The ﬁrst thing to notice is the simplicity of the claim. ψt ). that is assuming the existence of a constant r. but the measure that makes the discounted stock a martingale. we’ll price a call option — the right but not the obligation to buy a unit of stock for a predetermined amount at a particular exercise date. And to do that. The value (ST − k)+ only depends on the stock price at one point in time — namely the expiry time. the value of the replicating strategy (and thus the option) at time zero. T . such that dEt = φt dZt .
−1 (2) Form the process Et = EQ (BT XFt ).
& %
Call options
We should price something. if we denote the stock
. So when interest rates are nonzero. knowable by some time horizon T . (ST − k)+ . have associated replicating strategies (φt . 0). Or in more convenient notation. If we let this predetermined amount be k (in ﬁnancial terms. the strike of the option). And the arbitrage price of the claim is the expectation under Q of the discounted claim.
(3) Find a previsible process φt . what are the new rules? They are just discounted versions of the old rules:
' $
Three steps to replication (discounted case) (1) Find a measure Q under which the discounted stock price Zt is a martingale.
−1 where Q is the martingale measure for Bt − St . CONTINUOUS PROCESSES
$
Summary Suppose we have a BlackScholes model for a continuously tradable stock and bond. In addition.70 '
CHAPTER 3. Then all integrable claims X . Our formula tells us that this is given by
e−rT EQ (ST − k)+ . we can look at the process for St written in terms of the Q1 ˜ ˜ Brownian motion Wt . our claim is max(ST − k. the arbitrage price of such a claim X is given by
&
−1 Vt = Bt EQ (BT XFt ) = e−r(T −t) EQ (XFt ). So to ﬁnd the expectation of this claim we need only ﬁnd the marginal distribution of ST under Q. Since d(log St ) = σdWt + (r − 2 σ 2 )dt. say T . First we should ﬁnd V0 . then in formal notation.
15 What information about the drift was required?
Price dependence
For values of the current stock price s much smaller than the exercise price k .06. If we use the notation Φ(x) to denote (2π)− 2 −∞ exp −y 2 /2 dy the probability that a normal N (0. 2
we can write ST as se(Z+rT ) and thus the claim as the expectation e−rT E which equals
√ 1 2πσ 2 T
∞ log(k/s)−rT 1 x + 2 σ2T exp − 2σ 2 T 2
se(Z+rT ) − k
+
. So the marginal distribution for ST is given by s times the exponential of normal 1 with mean (r− 2 σ 2 )T and variance σ 2 T . signifying that the option is out of the money
.
se − ke
x
−rT
dx
This integral can be decomposed by a change of variables into a couple of standard 1 x cumulative normal integrals. what is the value of an option to buy the stock for $25 in two years time.
This is the BlackScholes formula for pricing European call options.8 BlackScholes in action
If a stock has a constant volatility of 18% and constant drift of 8%. r = 0. T ).8. given a current stock price of $20? The description ﬁts the BlackScholes conditions. σ = 0.221. we can calculate V0 as $1. k = 25. the right to sell a unit of stock for k . we have that log St = log s + σ Wt + (r − 2 σ 2 )t and thus 1 ˜ St = s exp σ Wt + (r − 2 σ 2 )t . (Put options. σ 2 T ). T ) = sΦ
− ke−rT Φ
. where BlackScholes formula
s 1 log k + r + 2 σ 2 T √ σ T 1 s log k + r − 2 σ 2 T √ σ T
V (s. with continuously compounded interest rates constant at 6%. and t = 2. by s. then we can calculate that V0 = V (s.
xxxx.14 Find the change of variable and thus prove the BlackScholes formula. 1) has value less than x. the value of the formula itself gets small.18.3. S0 . Thus using s = 20. BLACKSCHOLES IN ACTION
71
1 ˜ price at time zero. Thus if we let Z be a normal N (− 1 σ 2 T.
3. can be priced as a call less a forward — putcall parity.)
xxxx. TNxxx
Exercise 3. TNxxx
Exercise 3.
1. purchase a unit of stock for a strike price k at any time up to and including the expiration date T . and that choice can only use price information up to the present moment. which is the value of the corresponding forward if the option will be in the money and is zero otherwise. the option value gets larger. for values of s much greater than k . the option is worth more the more volatile the stock is. An option with almost inﬁnite time to maturity would have value approaching s. rather than only at that date. and becomes a forward. Conversely. At one extreme. CONTINUOUS PROCESSES
and unlikely to recover in time. At the other extreme.3. Exercise price k = $1. (s − k)+ . if σ is very large.14: Option price against stock price for times 3.72
CHAPTER 3. and 0. the option resembles a riskless bond and is just worth (s − ke−rT )+ . volatility σ = 1.14 that as the time to expiration gets closer to zero. The buyer of the option then has to make decisions from moment to moment to decide when and if to call the option. the curve gets closer to the option shape (s − k)+ . the option is worth s.
Time dependence
As the time to maturity T gets smaller. Correspondingly the option price is approximately s − ke−rT which is the current value of a stock forward struck at price k for time T . the chances of the price moving much more decreases and the option value gets closer and closer to the claim value taken at the current price.
American options
Sometimes an option has more optionality about it than just choosing between two alternatives at the maturity date. The buyer of an American call has the choice when to stop. say. if σ is very small.
Volatility dependence
All else being equal. as the cost now of price k is almost zero. the option loses most of its optionality. American options are the most wellknown examples of such derivatives.
Figure 3. It can be seen in ﬁgure 3. For larger times. Such a (random) time is called
. however. interest rate r = 0. giving the right to.
we have to prepare for the worst possible case. and receives a payoff Xa at time T . However the martingale representation theorem.8.
V0 = sup EQ e−rτ (Sτ − k)+ . to actually ﬁnd out how much stock would be required at each point of time to artiﬁcially construct the derivative. one is locally just a scaled version of the other. And if we have a restricted enough claim where the only input required from the ﬁltration for pricing the claim is the stock price at the current moment. that is. then the issuer’s hedge will produce a surplus by date T . Following a strategy which will result in exercising the option at the stopping time T . if we looked at the ratio of the change in the value of the option caused by a move in the stock price and the change in the stock price used. intuitively. BLACKSCHOLES IN ACTION
73
a stopping time.
As we do not know which τ will be used. one thing that would be useful to know would be the actual replicating strategy required.
τ
Pricing derivatives with optionality In general. If the purchaser does not exercise the option optimally. the theorem merely states that φt exists. at heart.3.
That hedge in full
Returning to the original European option. then the option issuer should charge
V0 = sup EQ e−rT Xa
a∈A
for it. this ought to be something like φt . φt . the cost at time zero of hedging that payoff is
EQ e−rτ (Sτ − k)+ .
. if the option purchaser has a set of options A. and charge the maximum value (maximized over all possible stopping strategies). then we could guess that the partial derivative of the option value with respect to the stock price is the φt we want. tells us that the reason that the discounted claim can be built from the discounted stock is that. the corresponding payoff is
(Sτ − k)+
at time τ. The amount of stock.
If the option issuer knew in advance which stopping time the investor will use. being martingales under the same measure. comes from the martingale representation theorem. Thus. after choosing a in A. The process φt is merely the ratio of volatilities. and moreover that the functional relation implied by this between the value of the claim and the current stock price is smooth. but unfortunately.
t)) = σSt ∂V ∂s ˜ dWt + rSt
2 ∂V ∂V 2∂ V + 1 σ 2 St + 2 2 ∂s ∂s ∂t
dt. this PDE. for some f .
. then in the value of the derivative at time t is equal to Vt = V (St . Using this substitution for φt and the fact that Vt = St φt + ψt Bt . The amount of stock held is the derivative of the value function with respect to stock price. t).
Because φ is always between zero and one. the inﬁnitesimal change in the value of the option.74
CHAPTER 3. coupled with the boundary condition that V (s.
Explicit BlackScholes hedge
The call option is a terminal value claim. we can also match the drift terms of the two SDEs to get a partial differential equation for V as
2 1 2 2∂ V σ s 2 ∂s2
+ rs
∂V ∂V − rV + = 0. Remem˜ bering that dSt = σSt dWt + rSt dt. Suppose the derivative X is a function of the terminal value of the stock price. where V (s. the option value is a wellbehaved function of the current stock price. T ) must equal f (s). For the oftenencountered case where the claim depends only on the terminal value. And since dBt = rBt dt . t). giving φt = ∂V . we need only ever have a bounded long position in the stock. Terminal value pricing If the derivative X equals f (ST ). Then the following is true. as described earlier. and so we can ﬁnd an expression for the hedge itself. Also the value of the bond holding at any time is
Bt ψt = −ke−r(T −t) Φ log St + r − 1 σ 2 (T − t) k √ 2 σ T −t . T − t) = Φ ∂s
1 log St + r + 2 σ 2 (T − t) k √ σ T −t
. In symbols
φt = ∂V (St . t) = exp − r(T − t) EQ f (ST )St = s
And then the trading strategy is given by φt =
∂V ∂s
(St . ∂s ∂t
Notoriously. The amount of stock in the replicating portfolio at any stage is the derivative ∂s of the option price with respect to the stock price.
But SDE representations are unique — so the volatility terms must match. t) is given by the formula
V (s. then Itˆ gives us o
dVt = d (V (St . so that X = f (ST ) for some function f (s).
Why? Consider dVt .
But we also know that dVt = φt dSt + ψt dBt from the selfﬁnancing condition. CONTINUOUS PROCESSES
And so it is. gives another way of solving the pricing equation. we have
˜ dVt = (σSt φt ) dWt + (rSt φt + rψt Bt ) dt.
We can calculate both the evolving worth of the option Vt and the amount of stock to be held. BLACKSCHOLES IN ACTION
75
which. signifying that the option will be exercised. the option becomes in the money and the option value moves like the stock price. to buy it at time T = 1 for the strike price of k = $12. In the case (A). to hedge the contract.3.8. then the stock holding grows to one unit and the value of the bond to −k . If the option is out of the money.
. Alternatively. then both the bond and the stock holding go to zero. Also the hedge gets closer and closer to one. reﬂecting the increasing worthlessness of the option. Below are two possible realizations of a stock price which starts at $10. Both are exponential Brownian motions with volatility 20% and growth drift of 15%. φt .17. assuming interest rates are 5%.
(a) Option value (A)
(b) Stock hedge (A)
Figure 3. that is the stock price is less than the exercise price. There are two possibilities as the time approaches maturity. In the case (B).16.15: Stock price (A) and (B)
Let us price an option on this stock. This combination exactly balances the now certain demand for a unit of stock in return for cash amount k .
(a) Stock price (A)
(b) Stock price (B)
Figure 3. these processes are shown in ﬁgure 3.16: Option value and Stock hedge of (A)
As time progresses. these processes are shown in Figure 3. is bounded by the exercise price k . although always a borrowing. if the price stays above the exercise value.
Example — hedging in continuous time
This can be seen operating in practice.
TNxxx
Exercise 3. with respect to a measure which makes the discounted stock a martingale.17: Option value and Stock hedge of (B)
This time the option is not exercised and both the value of it and the hedge go to zero over time. These three surprises conspire to make the result look easier to get at than perhaps it really is. Not something we had any right to expect. we can replicate any claim. CONTINUOUS PROCESSES
(a) Option value (B)
(b) Stock hedge (B)
Figure 3. changing to the martingale measure has a remarkably simple effect on the process St — only the drift changes. Moreover. to another constant value. even the volatility σ stays the same.16 A stock has current price $10 and moves as an exponential Brownian motion with upward drift of 15% a year (continuously compounded) and volatility of 20% a year. Current interest rates are constant at 5%. The replicating portfolio has a value given by the expected discounted claim. Before we push on.
.17 For the same stock. what is the value of a derivative which pays off $1 if the stock price is more than $10 in a year’s time?
Conclusions
Even with a respectable stochastic model for the stock. Something subtle and beautiful really is going on under all the formalism and the result only serves to obscure it.76
CHAPTER 3. stop and admire the view. The stock remains an exponential Brownian motion. What is the value of an option on the stock for $12 in a year’s time?
xxxx.
xxxx. TNxxx
Exercise 3.
As with stocks. equities and bonds — don’t actually ﬁt the simple asset class we devised. Even leaving aside the issues of transaction costs and illiquidity. holding the basic asset. currency. is a risky business. and C0 dollars buy a pound now.1 Foreign exchange
In the foreign exchange market. and
77
. But cash in both currencies attracts interest. though.
Forwards
Consider. a forward transaction: a dollar investor wanting to agree the cost in dollars of one pound at some future date T . The sophistication we have to peddle now is ﬁnancial. our cash holding wasn’t cash but a cash bond.Chapter 4
Pricing market securities
T
he BlackScholes model we have seen so far has a simple mathematical side but it has an even simpler ﬁnancial side. Just retreading the same mathematics for each of these will be enough to keep us busy. Let’s make things concrete. say. equities pay dividends.
4. the replicating strategy to guarantee the forward claim is static. And just as in the simple BlackScholes model. The asset we considered was a stock which could be held without additional cost or beneﬁt and was freely tradable at the price quoted. one pound sterling varies from moment to moment just as a US stock does. Even vanilla products — foreign exchange. At time t we
• own e−uT units of sterling cash bonds. Suppose the constant dollar interest rate is r. And with this risk comes demand for derivatives: claims based on the future value of one unit of currency in terms of another. Consider the following static replicating strategy. The dollar value of. We buy pounds now and sell dollars against them. so our cash holdings here will be cash bonds as well. like the stock market. Foreign exchange involves two assets which pay interest. the sterling interest rate is u. and bonds pay coupons. not much of the ﬁnancial market is like that.
yields two processes. but it’s a sterling price. but sterling cash isn’t a tradable instrument in our market. and Ct be the dollar worth of one pound. The process Ct represents the dollar value of one pound sterling. Then our model is Dollar bond Sterling bond Exchange rate
Bt = ert . then. PRICING MARKET SECURITIES
• go short C0 e−uT units of dollar cash bonds. and at time T the sterling holding will be one pound as required and the dollar short holding will be C0 e(r−u)T — the forward price we require. and that rate is u not zero. Bt and St . σ and µ. One is uncomplicated — the dollar bond is straightforwardly a dollar tradable much as the cash bond was in the basic account of BlackScholes. Ct = C0 exp(σWt + µt). Translation.
. that is by multiplication by Ct . as a tradable but it isn’t. But the other is not.
The dollar investor
The underlying ﬁnance dictates that there are two tradables available to the dollar investor. the exchange rate. u. And that makes a difference. the product of the two St = Ct Dt is a dollar tradable. Dt by itself isn’t a dollar tradable either — it is the price of a tradable instrument. which mirror the basic BlackScholes set up. We must be careful in extending our simple model to foreign exchange — both instruments now make payments.
At time zero the portfolio has nil value. our market will be: BlackScholes currency model We let Bt be the dollar cash bond. We would like to think of the stochastic process Ct . On the other hand.78
CHAPTER 4. the existence of the sterling cash bond Dt sets an interest rate for sterling cash by arbitrage. Contrast this with the stock forward price S0 erT . and the dollar value of the holding will be given by the translation of the sterling price Dt into dollars. Dt = eut .
for some Wt a PBrownian motion and constants r. Fortunately. The dollar investor can hold sterling cash bonds. Dt its sterling counterpart. To hold cash naked would be to set up an arbitrage against the cash bond — to put it another way.
BlackScholes currency model
There are three instruments and processes to model — two local currency cash bonds and the exchange rate itself. Following the mathematical simplicity of BlackScholes.
˜ Can we make this into a martingale under some new measure Q? Only if Wt = 1 Wt + σ −1 µ + u − r + 2 σ 2 t is a QBrownian motion. which is made possible as before by the CameronMartinGirsanov theorem. deﬁne the process Et to be the conditional expectation process EQ (BT XFt ).7. is a martingale. 2
Step two
−1 Given this Q. if dEt = φt dZt — which is precisely what the martingale representation theorem guarantees. FOREIGN EXCHANGE
'
79 $
Three steps to replication (foreign exchange) (1) Find a measure Q under which the sterling bond discounted by the dollar bond −1 −1 Zt = Bt St = Bt Ct Dt . which as noted before is a Qmartingale.4. such that
t
Et = E0 +
0
φs dZs . such that dEt = φt dZt . ψt ) detailing holdings of our two dollar tradables St and Bt . Then.
(3) Find a previsible process φ. so we try
• holding φt units of sterling cash bond. This portfolio is only selfﬁnancing if changes in its value are only due to changes in the assets’ prices. and • holding ψt = Et − φt Zt units of dollar cash bond. that is dVt = φt dSt + ψt dBt .
& %
Step one
The dollar discounted worth of the sterling bond is
Zt = C0 exp σWt + (µ + u − r)t .
Step three
The martingale representation theorem produces an F previsible process φt linking Et with Zt .
Now where? We need a replicating strategy (φt .
−1 (2) Form the process Et = EQ (BT XFt ).
The dollar value of the replicating portfolio at time t is Vt = φt St + ψt Bt = Bt Et . 2
and thus
˜ Ct = C0 exp σ Wt + (r − u − 1 σ 2 )t . or as was shown to be equivalent in section 3. under Q ˜ Zt = C0 exp σ Wt − 1 σ 2 t .
.1.
1) random variable. thus dEt = e−uT dZt . and so the required hedge φt is the constant e−uT .
That is. σ and k are constants.
Example — call option
A sterling call. our payoff at time T is
X = CT − k.
−1 The discounted portfolio value is Et = Bt Vt = e−uT Zt − e−uT C0 . and ET is the discounted claim BT X .
Example — forward contract
A sterling forward contract. and F . then ¯
E F exp σ Z − ¯
1 2 ¯ 2σ
−k
+
= FΦ
log F + 1 σ 2 k 2¯ σ ¯
− kΦ
1 ¯ log F − 2 σ 2 k σ ¯
. Suppose we have a contract which allows us the option of buying a pound at time T in the future for the price of k dollars. This conﬁrms our earlier intuition. With this strike. and ψ is the constant −e−uT C0 .80
CHAPTER 4. The dollar payoff at time T is
X = (CT − k)+ . Because CT is lognormally distributed we can evaluate this easily using a probabilistic result:
Lognormal call formula If Z is a normal N (0. PRICING MARKET SECURITIES
−1 Since VT = BT ET . the current price for sterling discounted by a factor depending on the difference between the interest rates of the two currencies. So the forward price at time zero for purchasing sterling at time T is k = EQ (CT ) or 1 ˜ F = EQ C0 exp σ WT + r − u − 2 σ 2 T
= e(r−u)T C0 .
. ψt ) which replicates our arbitrary claim X . At what price should we agree to trade sterling at a future date T ? If we agree to buy a unit of sterling for an amount k of dollars.
−1 The value of the payoff at time t is Vt = Bt EQ (BT XFt ).
Option price formula (foreign exchange) All claims have arbitrage prices and those prices are given by the portfolio value
−1 Vt = Bt EQ (BT XFt ). the contract’s value at time t is
Vt = e−uT eut Ct − ert C0 .
−1 Its worth at time t is Vt = Bt EQ (BT XFt ) which is e−r(T −t) EQ (CT − kFt ).
where Q is the measure under which the discounted asset Zt is a martingale. we have a selfﬁnancing strategy (φt .
we can follow again our threestep replication program. the value of CT can be written in the form F exp(¯ X − 1 σ 2 ).
The sterling investor
A sterling investor sees things differently. Were we operating in pounds we would not be wanting dollar price processes of tradable instruments but sterling ones. Ft = e(r−u)(T −t) Ct . The ﬁrst of these is simply the sterling bond Dt = eut . This has the value
−1 −1 Ct = C0 exp(−σWt − µt). if
1 ˜ Wt£ = Wt + σ −1 µ + u − r − 2 σ 2 t
is Q£ Brownian motion. any more than Ct was for the dollar investor.
ψt = −ke−rT Φ
1 log Ft − 2 σ 2 (T − t) k √ σ T −t
where Ft is the forward sterling price at time t. With our two sterling tradable prices.4. which will be our basic unit of account.
The hedge is
φt = e−uT Φ log Ft + 1 σ 2 (T − t) k √2 σ T −t .
where Q£ is the measure under which the sterling discounted asset Yt is a martingale. which the theorem ¯ ¯ tells us is
V0 = e−rT FΦ log F + 1 σ 2 T k √ 2 σ T − kΦ log F − 1 σ 2 T k √ 2 σ T .
This discounted price process Yt will be a martingale under the new measure Q£ . Then hedging will be possible as before. Option price formula (sterling investor) The value to the sterling investor of a sterling payoff X at time T is
−1 Ut = Dt EQ£ (DT XFt ). The sterling discounted value of the dollar bond is
−1 −1 −1 Yt = Dt Ct Bt = C0 exp(−σWt − (µ + u − r)t).
. There is also the inverse exchange rate process Ct−1 — the worth in pounds of one dollar.
but it is not the sterling price of a tradable instrument.1. FOREIGN EXCHANGE
81
As the forward price F is EQ (CT ). where σ 2 is the variance of log CT . . namely σ 2 T and Z is a normal σ ¯ 2¯ N (0. 1) under Q. + 1 The option price at time zero is then F exp σ Z − 2 σ 2 − k . Dt and Ct−1 Bt . Our other actual sterling tradable price process is the sterling value of the dollar bond Ct−1 Bt .
Ct Ut .
−1 EQ£ (XFt ) = ζt EQ (ζT XFt ).
Note that ζt is (up to a constant) the dollar discounted worth of the sterling bond. To the dollar investor.4. All investors. PRICING MARKET SECURITIES
Change of numeraire
A worrying possibility now surfaces — the measures Q and Q£ are different. whatever their currency of account. is the dollar worth of the sterling valuation. C0 ζt = Zt = Bt Ct Dt . the claim is worth at time t
−1 Vt = Bt EQ BT X Ft
dollars. So they agree not only on the prices but also on the hedging strategy. at time T . Recall also (RadonNikodym fact (ii) of section 3.
Do these two prices agree? That is.82
CHAPTER 4. the claim pays off CT X pounds. will agree on the current value of a derivative or other security. Similar calculations show that the dollar and sterling investors’ replicating strategies for X are identical.4) that for any random variable X which is known by time T . of changing numeraires are in section 6.
which is (substituting in the ζt expression) equal to
−1 Ct Ut = Bt EQ BT X Ft = Vt .
Thus the payoff of X dollars at time T is worth the same to either investor at any time beforehand. or lack of it. The difference of martingale measures only reﬂected the different numeraires of the two investors rather than any fundamental disagreement over prices. 2 dQ
The Qmartingale associated with the RadonNikodym derivative. formed by conditional expectation is
ζt = EQ dQ£ Ft dQ
1 ˜ = exp σ Wt − 2 σ 2 t .
. Further details on the effect. −1 Concretely.
−1 To the sterling investor. Will the dollar and sterling investors disagree about the price of the same security? Suppose X is a dollar claim which pays off at time T . the same as the original dollar valuation Vt ? ˜ ˜ The Q£ Brownian motion Wt£ is equal to Wt − σt. and its sterling worth at time t is −1 −1 Ut = Dt EQ£ DT (CT X) Ft
pounds. so that by the converse of the Cameron MartinGirsanov theorem the RadonNikodym derivative of Q£ with respect to Q (up to time T ) must be
dQ£ ˜ = exp σ WT − 1 σ 2 T . rather than X dollars.
So the dollar worth of the sterling investor’s valuation is
−1 −1 −1 −1 −1 Ct Ut = Ct Dt EQ£ DT CT X Ft = Ct Dt ζt EQ ζT DT CT X Ft .
that cash is immediately used to buy a little more stock. and the worth of the portfolio is
˜ St = S0 exp σWt + (µ + δ)t . what we bought is worth not just the price of the stock itself. the number of stock units held by the portfolio will be exp(δt). Assumptions are all. Just as with foreign exchange. That is. EQUITIES AND DIVIDENDS
83
4. which will purchase δdt more units of stock. Equity model with continuous dividends Let the stock price St follow a BlackScholes model. At time t. independent of the stock price. If we buy the stock for S0 . then we would have reinvested in the cash bond (for an example of this see section 4. It is simplest to begin with a dividend which is paid continuously.
Replicating strategies — equities
Our deﬁnition of a portfolio of stock and bond (φt . If we had assumed that the dividend stream was known in advance.
where δ is a constant of proportionality. but they can be modiﬁed to handle dividend payments. The process St is no longer the value of the asset as a whole. and to ﬁnd a new process as we did in foreign exchange. where φt = e−δt φt . which involves St but is a tradable.2.2 Equities and dividends
An equity is a stock which makes periodic cash payments to the current holder. namely St . we are continuously reinvesting the dividends in the stock. ψt ) can be rewritten as a portfolio ˜ ˜ of the reinvested stock and bond (φt . Consider the following simple portfolio strategy. Our previous models treated a stock as a pure asset. with value Vt = φt St + ˜ ˜ ψt Bt = φt St + ψt Bt . The dividend payment made in the time interval of length dt starting at time t is
δSt dt. ψt ). our problem is that the process St is not a tradable asset.4. We need to translate St somehow. but also the total accumulated dividends. because it is not enough. The portfolio starts with one unit of stock. by the time we come to sell it at time t. As a consequence it made it natural to construct the tradable by reinvesting in the stock.
Note how the structure of the model’s assumptions made the translation straightforward. We assumed that the dividend payments were a constant proportion of the stock price. The inﬁnitesimal payout is δSt dt per unit of stock.3 on bonds). which under the model will depend on all the different values that the stock has taken up until time t. costing S0 . The advantage of the new framework is that the selfﬁnancing
. St = S0 exp(σWt + ρt) and Bt be a constantrate cash bond Bt = exp(rt). and at every instant when the cash dividend is paid out.
That is. To construct a strategy to hedge a claim X maturing at date T .
The value of k which gives the contract initial nil value is the forward price of ST .
˜ ˜ ˜ ˜ The trading strategy is to hold φt units of the translated asset St and ψt = Et − φt Zt units of the cash bond.
which is lognormally distributed. Why? Again because of our assumption — if the dividend payments are a known proportion of the stochastic St . we have no choice but to hide them in the stock itself. PRICING MARKET SECURITIES
equation retains the familiar form
˜ ˜ dVt = φt dSt + ψt dBt . there exists a ˜ previsible process φt such that
−1 −1 Et = EQ (BT XFt ) = EQ (BT X) + t 0
˜ ˜ φs dZs . we are continually buying more with the dividend income. In terms of our original securities. Instead of simply holding a certain amount of stock until T .
˜ so that we want a measure Q under which Wt = Wt + σ −1 (µ + δ + 1 σ 2 − r)t is 2 ˜ ˜ ˜ Brownian motion. this equation would need to be modiﬁed by the dividend cash stream. That is.
Example — forward
An agreement to buy a unit of stock at time T for amount k has payoff
X = ST − k. under the martingale measure
1 ˜ St = S0 exp σ Wt + (r − δ − 2 σ 2 )t .
F = e(r−δ)T S0 . this amounts to holding ˜ φt = eδt φt units of the stock St and the same ψt units of the bond Bt . Note the slightly surprising dynamic strategy for the forward. changes in the portfolio value are due both to trading proﬁts and losses (the dSt and dBt terms) and also to dividend payments.
whereas in the plain stock/bond notation.
The hedge is then to hold φt = e−δ(T −t) units of the stock and ψt = −ke−rT units of the bond at time t. again we follow the simple BlackScholes model. dZt − σ Zt dWt . Now Zt has SDE
1 ˜ ˜ dZt = Zt σdWt + (µ + δ + 2 σ 2 − r)dt . and use the martingale representation theorem. becoming dVt = φt dSt + ψt dBt + φt δSt dt.
Its worth at time t is
Vt = EQ e−r(T −t) (ST − k) Ft = e−δ(T −t) St − e−r(T −t) k. Working now with our reinvested stock. So under this martingale measure Q.84
CHAPTER 4. as usual we want to make the discounted −1 ˜ ˜ ˜ asset Zt = Bt St into a martingale.
. Thus.
The forward price for ST is
F = e(r−δ)T S0 = 1. How much is this payout worth? Our data are FTSE drift µ = 7% FTSE volatility FTSE dividend yield
σ = 15% δ = 4%
UK interest rate r = 6.3.5% As FTSE is composed of 100 different stocks. we can value these calls (per unit) at 0.4. which equals
V0 = e
−rT
FΦ
log F + 1 σ 2 T k √ 2 σ T
− kΦ
log F − 1 σ 2 T k √ 2 σ T
. X is actually the difference of two FTSE calls (plus some cash).
.) Again the BlackScholes call option formula reemerges — if the martingale measure Q makes the process under study. More precisely. St .0183 — about 5% too high.9ST }.3e−rT + 0.133. we would incorrectly have valued the contract at 1.444)+ − (ST − 2)+ . but the dividends are not reﬂected in the index. exercised at time T has payoff X = (ST − k)+ . and value at time zero of V0 = EQ e−rT (ST − k)+ . (Here Φ(+) and Φ(−) refer respectively to the two Φ terms in the above equation.3 + 0. EQUITIES AND DIVIDENDS
85
Example — call option
A call struck at k . The claim X is
X = min max{1. their separate dividend payments will approximate a continuously paying stream. Knowing the forward F and the term volatility σ is enough to specify the price. with a guaranteed minimum payout and a maximum payout.9 (ST − 1.
where T is 5 years and the initial FTSE value S0 is 1. This claim can be rewritten as
X = 1. have a lognormal distribution.0422 − 0.
That is.0067) = 0.9(0. it is a ﬁveyear contract which pays out 90% times the ratio of the terminal and initial values of FTSE.0067 respectively.8 .2. 1.
Were we to have forgotten that the constituent stocks of FTSE pay dividends.
where F is the forward price above e(r−δ)T S0 . The hedge will be to hold e−gd(T −t) Φ(+) units of the stock and have a negative holding of ke−rT Φ(−) units of the bond. then the theorem in section 4.0422 and 0.9712. or 180% if otherwise it would be more.
Using the above call price formula for dividendpaying stocks.
Example — guaranteed equity proﬁts
A contract pays off according to gains of the UK FTSE stock index St . 0. The worth of X at time zero is then
V0 = 1.1 comes into play. Or it pays out 130% if otherwise it would be less.
. Translation. are known in advance. We face two problems. .
where n[t] = max{i. There is also a cash bond Bt = exp(rt). we can assume the dividend payout exactly equals the instantaneous decrease in the stock price. the current holder of the equity receives a payment of a fraction δ of the current stock price. At time t. but this presents no real problems for our basic model. We are back in familiar territory. every time the stock pays a dividend we reinvest the dividend by buying more stock. Such a strategy is equivalent ˜ to holding φt = (1 − δ)−n[t] φt units of the actual stock St ˜ ˜ ˜ ˜ The discounted value of the (φt . where φt is the number of units of St we hold at time t.. T2 . where
˜ St = (1 − δ)−n[t] St = S0 exp(σWt + µt). Let us assume that the times of dividend payments T1 . PRICING MARKET SECURITIES
Periodic dividends
In practice. . The second is more se1 rious. As dZt =
. ψt ) portfolio is Et = φt Zt + ψt . Thus St is discontinuous — it doesn’t ﬁt our deﬁnition of a stochastic process. The stock price must also instantaneously decrease by the same amount — or else there would be an arbitrage opportunity. ˜ ˜ As before. Away from the times Ti . the equity pays a dividend of a fraction δ of the stock price which was current just before the dividend is paid. Starting with one unit of stock. ψt ). The portfolio will be ˜ ˜ selfﬁnancing if dEt = φt dZt . The stock price process itself is modeled as
St = S0 (1 − δ)n[t] exp(σWt + µt). At any time T = Ti . Ti ≤ t} is the number of dividend payments made by time t. but at those times it has discontinuous jumps. .
Replicating strategy
˜ ˜ ˜ Our trading strategy will then be (φt . Fortunately. should provide a cure. . we want to ﬁnd a Q which makes Zt into a martingale. however. St has the usual SDE of dSt = St σdWt + (µ + 2 σ 2 )dt . and at each time Ti . Consider the following trading strategy. . T2 . and ψt is the amount of the cash bond Bt . ˜ we will have (1 − δ)−n[t] units of the stock.86
CHAPTER 4. St is tradable but our arbitrage justiﬁed assumption that the dividend pay˜ ments match the stock price jumps feeds through into making St continuous as well. ˜ As before. The ﬁrst is the familiar one that St is not by itself the price of tradable asset. then. and the value of our portfolio will be St . an individual stock pays dividends at regular intervals rather than continuously. where Zt is the −1 ˜ ˜ discounted value of the reinvested stock price Zt = Bt St . Equity model with periodic dividends At deterministic times T1 . translation cures this as well.
Were interest rates completely constant at rate r it would have present value at time t of e−r(T −t) . uncertainty about their future values would cause a discount bond price to move randomly as well. We might. The interplay of interest rates of different maturities and the arbitrage mineﬁeld that models have to tiptoe through is not something we want to worry about in a simple BlackScholes account. but then it can be argued that there are links between stock or foreign exchange prices and the cash bonds as well. Then Zt is also a Qmartingale. where X is the option on the stock
which we wish to hedge. but the shortterm interest rate will be deterministic. this will have no drift if Wt = Wt + σ −1 (µ + 1 σ 2 − r)t 2 ˜ is QBrownian motion.3. Bond prices will vary stochastically. will have to wait for chapter ﬁve and term structure models. As a consequence we will try to take a schizophrenic attitude to interest rates. is
St = S0 (1 − δ)n[t] eσWt +(r− 2 σ )t . A full model of discount bonds. BONDS
87
1 ˜ ˜ Zt σdWt + (µ + 2 σ 2 − r)dt . under Q. Over short time horizons most practitioners ignore these links in all three markets. or for that matter coupon bonds. In the real markets there is clearly a link. The stock price. And with varying interest rates.
˜
1 2
Since this is lognormal.4. ˜ Finally. however.3 Bonds
A pure discount bond is a security which pays off one unit at some future maturity time T . want to consider the effect of interest rates being stochastic — much as they are in real markets. the BlackScholes price for a call option struck at k is equal to
V0 = e
−rT
FΦ
log F + 1 σ 2 T k √ 2 σ T
− kΦ
log F − 1 σ 2 T k √ 2 σ T
. −1 We can form the process Et = EQ (BT XFt ). the martingale representation theorem produces a hedging process φt and ˜ ˜ the corresponding ψt can be set to be Et − φt Zt . and the value at time zero of the claim X is EQ (BT X). So hedging is still possible in this −1 case.
4. with the forward price for ST equal to F = S0 (1 − δ)n[T ] erT .
Discount bonds
The BlackScholes model for discount bonds is:
.
this model is indistinguishable from the simple BlackScholes model for stocks. Tn before maturity. T . Thus the forward price for purchasing the bond at time T < τ is
F = EQ (ST ).
for all times t less than T . . PRICING MARKET SECURITIES
Discount bond model We assume a cash bond Bt = exp(rt) for some positive constant r. thus its price at time τ must be Sτ = 1. . The simplest model is to view coupons that occur before time T as coming under the regime of the deterministic cash bond. of log ST (σ is the term volatility).
where Q is the measure under which e−rt St is a martingale. Here the schizophrenia extends to the treatment of coupons before and after the expiry date. of the option. the amount of the coupon is known in advance. and coupons occurring after time T (including the redemption payment at maturity) as following a stochastic price process. but unlike the equity model. then the price of a call on ST struck at k is
e−rT FΦ log F + 1 σ 2 T k √ 2 σ T − kΦ
1 log F − 2 σ 2 T k √ σ T
. the drift and volatility will conspire to ensure this pull to par — and indeed this will happen in chapter ﬁve. under Q. Since σ 2 T is the variance. Not only does the distinction between the deterministic cash bond and the stochastic discount bond get harder to maintain as T approaches τ . and a discount bond St whose price follows
St = S0 exp(σWt + µt). . The bond promises one unit at time τ . In a good model. . with our assumption that T is much before the maturity τ .
We have to be careful. but for similar reasons it gets harder to justify a simple drift µ and a constant positive σ .
Bonds with coupons
Most market bonds do not just pay off one unit at maturity. Such coupon payments may resemble dividend payments. In formulation. T2 . though.88
CHAPTER 4. Here if we let T = τ . some time horizon T long before the maturity time τ of the bond. we would have no such guarantee. but also pay off a series of smaller amounts c at various predetermined times T1 .
.
4. and µ. The value of X at time zero must now be −1 EQ (BT X). ψt ) a hedging strategy for X . Denoting I(t) = min{i : t < Ti } to be the sequence number of the next coupon payment after time t. Ti < T ) to be worth ce−r(Ti −t) at time t (t < Ti ). where
j
˜ St =
i=1
ce−r(Ti −t) + A exp(σWt + µt). and for the sum of all the postT payments to evolve as an exponential Brownian motion A exp(σWt + µt).3. and Et be its discounted value Et = φt Zt + ψt . we want to make Zt into a martingale by changing measure. where φt is the amount of the asset St held at time t.
Replicating strategy
˜ ˜ ˜ We describe a portfolio as (φt . the total number of payments before time T . up to a horizon τ . We let Vt be the value of the ˜ ˜ ˜ ˜ ˜ portfolio. The portfolio is selfﬁnancing if dEt = φt dZt . in the cash bond. as they occur.
Speciﬁcally. . BONDS
89
Coupon bond model There is a simple cash bond Bt = exp(rt). ψt ). for example. and ψt is the direct holding of the cash bond Bt = ert . This will be a Qmartingale if Wt = Wt + σ −1 (µ + 2 σ 2 − r)t is QBrownian motion. this time we manufacture a continuous tradable asset by holding one unit of the coupon bond and investing all the coupon ˜ payments.
. where Zt ˜ ˜ ˜ is the discounted value of the asset St . Vt = φt St + ψt Bt .. and a coupon bond which pays off an amount c at times T1 . then the price of the bond at time t is then
j
St =
i=I(t)
ce−r(Ti −t) + A exp(σWt + µt). we model the ﬁrst sort of coupon (payable at. for constants A. the process Et = EQ (BT XFt ) can be repre˜ ˜ ˜ ˜ ˜ sented as dEt = φt dZt for some previsible process φt . But because the coupon payments are known in advance.
This is now a tradable asset with a continuous stochastic process. The value of this asset is St . and j to be I(T ) − 1. We can set ψt to be Et − φt Zt . and again we can use a translation rather like the one used for equity dividends (section 4. Again St is discontinuous at the coupon payment times. −1 For an option X payable at time T . . for t < T.2). ˜ As usual. . σ . ˜ so that (φt . T2 . In fact ˜ Zt is just a constant cash sum of j ce−rTi plus an exponential Brownian motion i=1 1 ˜ A exp(σWt + (µ − r)t).
t < T.
the model process had to have dividends recombined to make it tradable.
Martingales are tradables
Suppose that there is some measure Q under which the discounted tradable.
This is lognormally distributed. ψt ) where at time t we are
• long φt of the tradable St .90
CHAPTER 4.4 Market price of risk
Now is the time to tie some loose ends together.
Taking our cue from all the examples so far. we could create a portfolio (φt .1 to see that the forward price for ST is F = AerT and the value of a call on ST struck at k is
e
−rT
FΦ
log F + 1 σ 2 T k √ 2 σ T
− kΦ
1 log F − 2 σ 2 T k √ σ T
. Some of the tradable/nontradable distinction is going to have to be founded on goodwill. But if we decide on a particular process St representing something truly tradable and select an appropriate discounting process Bt . the martingale representation theorem gives us that. But the distinction has so far been a common sense one — can we do any better? To some extent.
4. After all whether something can be traded or not in a free market is not a mathematical decision. the coupons had to be reinvested in the numeraire process. Zt = −1 Bt St . And for bonds. PRICING MARKET SECURITIES
Under Q. The foreign exchange process had to be converted from a nontradable cash process to a tradable discount bond process.
. The same pattern has been repeating through all the examples so far — the stochastic processes we have been using as models in this chapter have been tied to tradable quantities only indirectly. what can we say about another process Vt adapted to the −1 same ﬁltration Ft such that Et = Bt Vt is also a Qmartingale? Firstly. is a Qmartingale. For equities. Underlying all this was a tradable/nontradable distinction — we couldn’t use the martingale representation theorem to replicate claims until we had something tradable to replicate with. as long as Zt has nonzero volatility. the price of the bond at time T is just
1 ˜ ST = A exp σ WT + (r − 2 σ 2 )T . then we can explore the market they create. • long ψt = Et − φt Zt of the tradable Bt . we can ﬁnd an F previsible process φt such that
dEt = φt dZt . so we can follow the call formula from section 4. yes.
Within an established (complete) market of tradable securities. there is only room for two ‘independent’ tradables in a market deﬁned by
. It is tradable if its discounted price is a martingale under the martingale measure Q. ψt ) portfolio are explainable in terms of changes in value of the tradable constituents alone. So it seems reasonable enough to ennoble Vt with the title tradable as well. we could buy unlimited amounts of V and sell unlimited amounts of U collecting the cash up front. Bt St . Us were greater than Vs . the process is just part of the ‘linear span’ of St and Bt .4. we would run the engine in reverse. then. If. Thus if Vt were genuinely tradable. MARKET PRICE OF RISK
91
Then as before we can show that (φt .4. Being a Qmartingale after discounting is enough to ensure that it can be made costlessly from tradables — so it might as well be tradable itself. So we have two tradables. Then from our deﬁnition of a martingale. In other words we can make Vt out of St and Bt . Ut and Vt .
Nonmartingales are nontradables
−1 What about the other way round? Suppose Bt Vt was not a Qmartingale. We then have an arbitrage engine. ψt ) is a selfﬁnancing strategy. that is changes in the value of the (φt . the other way round. Of course all the derivatives that we have been constructing out of −1 claims have this property — EQ (BT XFt ) is always a Qmartingale. it had better not be tradable. Then the terminal value of UT −1 will be equal to VT but at time s. To −1 avoid arbitrage engines. and is not tradable if it isn’t.
Tradable securities Given a numeraire Bt and a tradable asset St . The V − U portfolio can be sold for nothing at time T . We have something akin to a deﬁnition then. there must be a positive probability at some times T −1 −1 and s that EQ (BT VT Fs ) = Bs Vs . will be (possibly) different. such that they are identical at time T but different at some earlier time s (with positive probability). a process Vt represents a tradable −1 asset if and only if its discounted value Bt Vt is actually a Qmartingale. is a martingale. One way round. the market formed by St . if Bt Vt were not a Qmartingale. Bt and Vt would contain arbitrage opportunities — something we might want to dismiss by ﬁat. there is a straightforward way of checking whether another process is a tradable security or not. leaving just the (invested) cash as a guaranteed proﬁt. Us and Vs . What would happen if Vt were tradable and the market stumbled into this possible ﬁltration? Suppose we deﬁne another process Ut by simply setting Ut to be the cost of repli−1 cating the claim VT . And the value of this portfolio at time t is always exactly Vt . As Bt Ut is a Qmartingale we can view Ut as tradable by dint of being able to construct it from St and Bt . that is Ut = Bt EQ (BT VT Ft ). where Q −1 is the measure under which the discounted asset. say. And if Us were less than Vs .
1 If St is a tradable BlackScholes stock price under the martin˜ gale measure Q. 2. Both forms can be equally used to deﬁne such geometric Brownian motions. to deﬁne price processes by means of their SDEs. That is if
µ1 − r µ2 − r = . which can be thought of as just a change of notation. typically
dSt = St (σdWt + µdt). That model had stock price St = S0 exp(σWt + µt). with cash bond Bt = exp(rt). but the SDE formulation allows a greater general class of models to be more easily considered.
Tradables and the market price of risk
The market price of risk is best introduced through a slight modiﬁcation of the simple BlackScholes model. where α = 2r/σ 2 . we have that
˜ Wt = Wt +
µi − r σi
t
must be a QBrownian motion for i equal to 1 and 2. 2 show that
2 (i) Xt = St is nontradeable.
xxxx. then
γ= µ−r σ
. −α (ii) Xt = St . St = exp σ Wt + (r − 1 σ 2 )t .92
CHAPTER 4. PRICING MARKET SECURITIES
onedimensional Brownian motion — any more and there can be arbitrage. both in the same market — that is both are functions of the same Brownian motion Wt . we want the discounted prices of St and St to be martingales under the same measure Q.
which has solution St = S0 exp σWt + (µ + 1 σ 2 )t . So assuming a simple numeraire Bt = exp(rt). and SDE
1 dSt = St σdWt + (µ + 2 σ 2 )dt . σ1 σ2
In one of those coincidences that cause confusion.
i = 1. The only difference between 2 1 2 these two approaches is the subtraction of 2 σ from the drift. economists attach a meaning to this quantity — if we interpret µ as the growth rate of the tradable.
i i dSt = St (σi dWt + µi dt). TNxxx
Exercise 4. 1 2 Suppose then that we have a couple of tradable risky securities St and St .
1 2 Following the discussion on tradables. r as the growth rate of the riskless bond and a as a measure of the risk of the asset. and both are deﬁned via their SDEs. however. But this can only happen if the two changes of drift are the same.
We will ﬁnd it convenient. is tradable.
The market price of risk is only a convenient algebraic form for the change of measure from P to Q. not a new argument for using it. If we write the SDEs in terms of the QBrownian motion Wt :
˜ dSt = St σt dWt + µt dt . This translates naturally in SDE terms to sharing a market price of risk — the market price of risk is actually the drift change of the underlying Brownian motion given by CameronMartinGirsanov. in fact. Then deﬁning
γt =
µt − r σt
gives a time and state dependent market price of risk. So we have a natural means for sorting through SDEs for tradables. All tradable securities must instantaneously have the same market price of risk. Rigor will have to wait until section 6. independent of their riskiness σt — the measure Q is neutral with respect to risk.4. Using this language then gives us a simple and compelling categorization of tradables in terms of their SDEs — all tradables in a market should have the same market price of risk. They share a market price of risk not for profound economic reasons or because investors behave with certain risk preferences but for the reason that to do otherwise would produce a nonmartingale process with a consequent opportunity for arbitrage.4. the same as above will hold.
The general market price of risk
We can. generalize to more sophisticated onefactor models. a martingale measure with St .
where σt and µt are previsible processes. All tradables then have the same growth rate under Q as the cash bond.
The riskneutral measure
It is worth reﬂecting on what we have done — we have provided justiﬁcation that to be tradable in a market deﬁned by a stock St and a numeraire Bt is to share. Despite this variation. MARKET PRICE OF RISK
93
is the rate of extra return (above the riskfree rate) per unit of risk.1. but for now we can observe that a general stochastic price process St will have SDE
dSt = St (σt dWt + µt dt). We also have a natural explanation for the market terminology of Q as the risk˜ neutral measure. But we should not overstretch the economic analogy — within our one factor market all tradables are instantaneously perfectly correlated. ˜
then St is tradable if and only if its market price of risk is zero. As such it is often called the market price of risk.
. after discounting by Bt .
Yt = f (Xt ). is tradable.94
CHAPTER 4. they can be priced via the normal expectation route. then f is the exponential function f (x) = ex . PRICING MARKET SECURITIES
Nontradable quantities
But convenient it is. we might have a nontradable Xt which is modeled with the stochastic differential
dXt = σt dWt + µt dt.
. we can write down X ’s behavior under Q as
1 2 ˜ t + rf (Xt ) − 2 σt f (Xt ) dt.
where σt and µt are previsible processes and Wt is PBrownian motion. Then by Itˆ ’s formula. using this riskneutral SDE for Xt .
Examples
(i) If Xt is the logarithm of a tradable asset. Let’s return to our underlying theme — dealing with nontradable processes. 2
As Yt is tradable. Assuming the discount rate is constant at r. dXt = σt dW f (Xt )
Thus if we have claims on Xt . We have Xt nontradable but a deterministic function of Xt . but they needn’t be. equities and bonds we had a model for a process that had a ﬁxed relationship to a tradable but was itself nontradable. Y has differential increment o
2 dYt = σt f (Xt )dWt + µt f (Xt ) + 1 σt f (Xt ) dt. then the market price of risk for tradables is
γt = µ + 1 σ2 − r 2 . we can write down the market price of risk for Yt immediately. σ
and the corresponding riskneutral SDE for Xt is
1 ˜ dXt = σdWt + r − 2 σ 2 dt. Concretely. With foreign exchange.
γt =
2 µt f (Xt ) + 1 σt f (Xt ) − rf (Xt ) 2 σt f (Xt )
Since this market price of risk is simply the change of measure from P to Q. Here σt and µt might be constants or constant multiples of Xt . In the simple case where σt = σ and µt = µ are constants (the basic BlackScholes model).
that is Yt = f (Xt . The function f is thus chosen to be f (x. t) = xeδt .
The asset Yt = exp(δt)Xt made from instantaneously reinvesting the dividends back into the stock holding is a tradable asset. t) 2 dt. t). has a sterling denominated stock price. t) − 1 σt f (Xt . suppose interest rates follow the process rt . t)
where f and f are derivatives of f with respect to x.5. t) = eut /x tells us that its riskneutral SDE is
2 ˜ dXt = Xt σt dWt + (σt + u − r)dt . (ii) The price process St pays dividends at rate δSt . f (Xt .
4. or more strictly eut /Ct is a dollar tradable asset if German interest rates are constant at rate u.5 Quantos
British Petroleum. then the rate Ct paid in dollars is nontradable.
and thus the riskneutral SDE for Xt becomes
(iii) Foreign exchange. and ∂t f is the derivative of f with respect to t. if the current stock price were
.45. if Ct is equal to DM 1. the process worth $1. we could also consider it as a pure number which could be denominated in any currency. Xt is nontradable with stochastic differential
dXt = σt dWt + µt dt. Contracts like this which pay off in the ‘wrong’ currency are quantos. = δt σ σXt e ˜ dXt = Xt σdWt + (r − δ)dt . For instance.) However the process 1/Ct is tradable.
then the timedependent transform of f (x. t) − ∂t f (Xt .4.45 is not tradable. QUANTOS
95
Timedependent transforms More generally. The market price of risk for tradables is then
γt = µXt eδt + δXt eδt − rXt eδt µ+δ−r . (That is. the ‘wrong way round’. But instead of thinking of that stock price just in pounds. X has differential
˜ dXt = σt dWt +
2 rt f (Xt . Then under the martingale measure Q. a UK company. Let Ct be the dollar/mark exchange rate (worth in deutschmarks of one dollar). If Xt = Ct has SDE
dXt = Xt (σt dWt + µt dt). Let Xt be the process St and assume that it follows the BlackScholes model
dXt = Xt (σdWt + µdt).
and Y is a tradable security which is a deterministic function of X and time.
Before we tease out the tradable instruments in dollars. in dollars. whilst leaving the actual number unaltered.20. The payoffs we describe involve a nontradable quantity. the vector random variable (log St . then ρW1 (t) + 1 − ρ2 W2 (t) is also a Brownian motion. a simple derivative is given the added twist of paying off in a currency other than in which the underlying security is denominated. Quantos are best described with examples. namely ρ = 1 − ρ2 . We have not actually met multifactor models yet. Ct = C0 exp ρσ2 W1 (t) + ρσ2 W2 (t) + νt . ¯
where ρ is the orthogonal complement of ρ. This
is not the same as the worth of the BP stock in dollars — that would depend on the exchange rate. note the covariance of St and Ct . for some positive constant interest rates r and u. ¯ ¯ In addition there is a dollar cash bond Bt = exp(rt) and a sterling cash bond Dt = exp(ut). namely receiving the BP stock price at time T as if it were
in dollars in exchange for paying a preagreed dollar amount. And our intuition should warn us that this act of switching currency is not a foreign exchange quibble but something more fundamental. For the construction.20. PRICING MARKET SECURITIES
£5.3). This is a useful way to manufacture two Brownian motions which are correlated out of a pair which are independent. and σ2 . but it is not a traded security. The British Petroleum stock price in dollars is a meaningful concept.
• an option to receive the BP stock price less a strike price. and it has correlation ρ with the original Brownian motion W1 (t). the quanto model is: Quanto model The sterling stock price St and the value of one pound in dollars Ct follow the processes
St = S0 exp σ1 W1 (t) + µt . Our two random processes will be the stock price and the exchange rate. but they are no more problematic than singlefactor ones if we keep our head. that is $5. Here are three:
• a forward contract.96
CHAPTER 4. and a correlation ρ lying between −1 and 1.
• a digital contract which pays one dollar at time T if the then BP stock price is
larger than some preagreed strike. What we have done is a purely formal change of units.2: for ρ lying between −1 and 1. which will be driven by two independent Brownian motions W1 (t) and W2 (t). Suppose we have a simple twofactor model.
In each case. Given these constants. we could have a derivative that paid this price in dollars. positive volatilities σ1 . it is helpful to recall exercise 3. We suppose there exist the following constants: drifts µ and ν . Rigor can be found in the multiple stock models section (6. If we write our model in vector form. log Ct )
.
. and 1 is the constant vector (1. Ct Dt . γ2 (t)).3. or equivalently a market price of risk that represents this change of drift. Their SDEs are
2 dYt = Yt ρσ2 dW1 (t) + ρσ2 dW2 (t) + (ν + 1 σ2 + u − r)dt . ¯ ˜ 2
. ¯ 2 2 1 2 dZt = Zt (σ1 + ρσ2 )dW1 (t) + ρσ2 dW2 (t) + (µ + ν + 2 σ1 + ρσ1 σ2 + 1 σ2 − r)dt . Writing down the ﬁrst two of these tradables after discounting by the third. a constant volatility for Ct of σ2 and a correlation between them of ρ. the dollar worth of the stock. the dollar cash bond Bt . 2 2 ˜ Ct = C0 exp ρσ2 W1 (t) + ρσ2 W2 (t) + (r − u − 1 σ2 )t . . we know we want to ﬁnd a change of measure to make these martingales.
Tradables
What are the dollar tradables? Following the intuition of the foreign exchange section (4. Ct St .5. we have Yt = Bt Ct Dt and Zt = Bt Ct St respectively. W1 (t) and W2 (t). γ2 (t)). We want to choose these γ so that the drift terms in dYt and dZt vanish simultaneously. ¯ 2
(This can be checked using the nfactor Itˆ ’s formula of section 6. Here then we have a market price of risk γt = (γ1 (t).) o As in the market price of risk section. QUANTOS
97
is jointlynormally distributed with mean vector (log S0 + µt . σ1
and
γ2 =
2 ν + 1 σ2 + u − r − ρσ2 γ1 2 ρσ2 ¯
Thus under Q we can write the original processes St and Ct as
2 ˜ St = S0 exp σ1 W1 (t) + (u − ρσ1 σ2 − 1 σ1 )t . log C0 + νt ) and covariance matrix
σ1 0 ρσ2 ρσ2 ¯ t 0 0 t σ1 0 ρσ2 ρσ2 ¯ =
2 σ1 ρσ1 σ2 t 2 ρσ1 σ2 σ2
That is. In other words the market price of risk will be a vector (γ1 (t).4. Not surprisingly this means solving a pair of simultaneous equations. γ1 (t) will be the price of W1 (t)risk and γ2 (t) will be the price of W2 (t)risk. .1). we have ensured a constant volatility for St of σ1 . there are three: the dollar worth of the sterling bond. given by
γ1 =
2 µ + 1 σ1 + ρσ1 σ2 − u 2 . More details are in section 6. 1). . As there are two sources of risk. µ is their drift vector. there will be two separate prices of risk.3. and a dollar numeraire. the −1 −1 numeraire. Respectively. 1 2 2 µ + ν + 1 σ1 + ρσ1 σ2 + 2 σ2 − r 2
This is a particular case of the more general result that the multidimensional market price of risk is γt = Σ−1 (µ − r1) where Σ is the assets’ volatility matrix. or equivalently performing the matrix inversion
γ1 (t) γ2 (t) = ρσ2 ρσ2 ¯ σ1 + ρσ2 ρσ2 ¯
−1 2 ν + 1 σ2 + u − r 2 .
For this to be on market. TNxxx
Exercise 4. This actually makes some sense.2 Verify that the measure Q which has Brownian motions t ˜ Wi (t) = Wi (t) + 0 γi (s)ds (i = 1.98
CHAPTER 4. that is to have a value of zero. There isn’t a portfolio which is always worth a dollar amount numerically equal to the BP stock price. Then the value of the forward at time zero in dollars is
V0 = e−rT EQ (ST − k) = e−rT exp(−ρσ1 σ2 T )F − k .1. namely (u − r)/σ1 σ2 ) this stops the dollar discounted stock price being a Qmartingale and thus prevents the price in
dollars from being tradable. As σ1 and σ2 are both positive. given that
˜ ρW1 (t) + ρW2 (t) is another QBrownian motion (as was proved in exercise 3. 2) really is the martingale measure for Yt and Zt . And that’s precisely what our intuition warned us. it is clear that this quanto forward price is greater than the simple forward price if and only if the correlation between the stock and the exchange rate is negative. PRICING MARKET SECURITIES
xxxx. we can price up our quanto options. The drift has an extra term: −ρσ1 σ2 . This is not the same as the simple forward price F for sterling purchase. and Z is a normal N (0. this static replicating strategy
. we must set k to be F exp(−ρσ1 σ2 T ). Suppose we assumed that the quanto forward price was actually the same as the simple forward price F .
Forward
To price the forward contract. ¯˜ But the stock price St is different from what we expected. For every value of ρ (except one.
If our assumption about the quanto forward price also being F were correct then this portfolio would be costless at time zero.
Pricing
Since we have a measure Q.2). Reassuringly the exchange rate process is as it was in section 4. • short one unit of the simple sterling forward also struck at F . 2
where F is the local currency forward price of ST . then we could construct the following portfolio at time zero: by going
• long C0 exp (r − u)T units of the quanto forward struck at F . F = S0 euT . At time T . under which the dollar tradables are martingales. it helps to reexpress the stock price at date T as
√ 2 ST = exp(−ρσ1 σ2 T )F exp σ1 T Z − 1 σ1 T . 1) random variable under Q.
except that the stock St is the price in yen of NTT. then the value of this portfolio is positive. QUANTOS
99
would yield (in dollars)
C0 exp (r − u)T (ST − F ) − CT (ST − F ) = C0 exp (r − u)T − CT (ST − F ).
Call option
Finally. a Japanese stock. has price V0 = e−rT Q (ST > k). between the sterling and yen cases. For negative ρ the quanto forward must be greater than F . and ρ is their correlation. price options. And replication involves the exchange rate. If the stock price ends up above its forward and the FX rate is below its forward.
Noting that C0 exp (r − u)T is the forward FX rate for CT .
xxxx. but again replicating strategies. What is the one difference.
Perhaps not surprisingly for a lognormal model.
Again the surprise of the exp(−ρσ1 σ2 T ) term. Indeed it is. or if we write FQ = F exp(−ρσ1 σ2 T ) the quanto forward price.
Digital
Our digital contract.3 Suppose everything remains the same. If the quanto forward price really were F under these circumstances it wouldn’t be hard to construct an arbitrage. TNxxx
Exercise 4. this is just the original BlackScholes formula with the quanto forward. which is correlated with the stock price. I(ST > k) in dollars. Ct is the dollar/yen exchange rate (the worth in yen of one dollar). Surely the event of ST being greater than k is independent of whether the option is denominated in sterling or dollars. we can compute the option price of e−rT EQ (ST − k)+ as
V0 = e
−rT
FQ Φ
2 log FQ + 1 σ1 T k √2 σ1 T
− kΦ
2 log FQ − 1 σ1 T k √2 σ1 T
. in the expression for the quanto forward price?
. And in a ‘cleaner’ option. Negative correlation makes these winwin situations more likely — perfect negative correlation makes them inevitable. not expectation under P.4. then the value is also positive. consider the effect of negative correlation. then
V0 = e
−rT
Φ
2 log FQ − 1 σ1 T k √2 σ1 T
. And if the stock price ends up below F and the FX rate is above its forward.5.
interest rate swaps. money in ten years? Previously we made the modelling assumption that the cost of money is constant. Fortunately we shall still be able to calculate the prices of these contracts on exactly the same riskfree hedging basis as before. options on bonds. which records when we are to receive the later payment. And a dollar tomorrow is better than a dollar next year. are all derived from basic interestrate securities. but this isn’t actually so. In nominal cash terms. This basic contract only requires two numbers to describe it — its length. But just how much is that time worth — is every day worth the same or will the price of money change from time to time? The interest rate market is where the price of money is set — how much does it cost to have money tomorrow.Chapter 5
Interest rates
T
ime is money. not the interest rate market. exotic contracts on the time value of different currencies. Bonds. Matters such as creditworthiness and the like are not our concern here. to price them. In general. A dollar today is better than a dollar tomorrow. money behaves just like a stock with a noisy price driven by a Brownian motion.
5. We are solely concerned with the time value of money for defaultfree borrowing. money in a year. We can call the maturity date on which
100
. the market for such interestrate derivatives far outstrips that for stock market products. the worth of such a contract will depend on factors other than just the time value of money. just as stock options are derived from stocks in the market. and the ratio of the size of that payment to our initial payment. or maturity. The uncertainty of the market opens up the possibility of derivative instruments based around the future value of money. such as the credibility of the promisor and the perceived legality of the promise.1 The interest rate market
The most basic interest rate contract is an agreement to pay some money now in exchange for a promise of receiving a (usually) larger sum later. In this way. and it is for the bond market. The price of money over a term depends not only on the length of the term. but also on the momenttomoment random ﬂuctuations of the interest rate market.
1 is not the price process of an asset. Each bond cannot just be treated in isolation as if it were a stock. Generally the more distant the payment maturity date. the value at time t of receiving a dollar at time T . say.5. and to price options on this T bond by trading in it to hedge them. This matters because the bonds. The tenyear bond. Bond prices are thus a function of two time variables. is a process in time — the price process of a tradable security. quantifying exactly how much better it is to have cash now rather than later. but it won’t turn out to be a problem. which will have some worth at time t before T . that is P (T. We could have written the contract for any one of the unlimited number of possible maturity dates. and the fraction of the ﬁnal payment which is the initial. This price P (t.2: Price of 10year bond vs time
Figure 5. will be correlated. We feel we should be able to model its behavior. although different. P (0. the situation is much like the stock market in that here we have a tradable asset which has a stochastic price process. one dollar at time T can be bought at time zero for P (0. T ) dollars. T ) is its price at time zero.1.2 is the price of one particular asset (the tenyear discount bond). T ). Illustrated are sections along the lines t = 0 (ﬁgure 5. but a graph of the current price of a whole spectrum of different assets (the bonds of different maturities). This is the real challenge of the interest rate market: the basic discount bonds are parameterized by two time indices. In other words. the less the current worth of the bond.1) and T = 10 (ﬁgure 5. which we can explore by taking twodimensional graphical sections through it. For any one maturity T .2). The bond price graph is actually a twodimensional surface lying in threedimensional space. This reﬂects the current time value of money.
Discount bonds
But we can also regard the promise of a dollar as an asset.) But we haven’t got just one maturity. Stocks don’t do things like that. we have a noisy stochastic
. Now instead of a smooth graph. say.1: Bond prices now
Figure 5. and the nineyear bond are going to move in very similar ways in the short term. as stocks were.
Figure 5. P (t. rather than just one. T ) = 1. T ). (The only difference is the technical point that the bond evolves towards a known value — at time T the bond is worth exactly one dollar. Figure 5. and the price P (0. But it can have a different price at any other time t up to maturity T . which determine both the start of the contract and its end. THE INTEREST RATE MARKET
101
we are paid T . T ). This asset is called a discount bond. call it.
Other than saying now is better than later. due to increased uncertainty about fardistant interest
. the yield R(t. The difference in yields at different maturities reﬂects market beliefs about future interest rates. In this particular case. being the common value P (0. INTEREST RATES
process. Yield Given a discount bond price P (t. A more informative measure of the market is an indication of the implied average interest rate offered by a bond. Yield curves. we can produce a yield curve. Typically the yield does increase with maturity. T ) against T for some ﬁxed t.1. and the mapping from price to yield is onetoone for t less than T — no information is lost. T ) = − log P (t. revealing the average return of bonds stripped of the crude effects of maturity — the term structure of the market. the translation is friendlier to the eye. is called the yield. T ). that is.3: Yield curve at t = 0
Figure 5. r can be recovered from the price P (t. If there is a possibility that rates might be higher in the future. Interest rates are not constant. T )/(T − t). so the downwards slope of the price curve is inevitable. it doesn’t tell us very much on quick inspection.
Yields
The picture in ﬁgure 5.1 is not particularly sensitive to what the market is doing. R(t.4: Yield curve at t = 4
While the discount bond price curve contains exactly the same information as the yield curve. 10).102
CHAPTER 5. T ) is given by
R(t. or the worth now of receiving a dollar in ten years’ time. If interest rates were constant at rate r. The start point of this graph is the end point of ﬁgure 5. but that doesn’t stop us viewing this translation as potentially useful. the price of the T bond at time t would be e−r(T −t) . The rate we derive. a graph of R(t. T −t
Thus for any given discount bond price curve. T ) at time t.
Figure 5. thus redundant. T ) via the formula r = − log P (t. longterm loans will have to charge a higher rate than shortterm ones. up until it hits the value one at its maturity time. T ) . Long dated bonds always have lower prices. can be increasing or decreasing functions of T . on the other hand.
Figure 5. That is. In one instance. Interestingly though. ∆t
For ever smaller time increments this value more closely approximates to R(t. borrowing which is paid back (nearly) instantly. t). which is given by both the expressions and
rt = R(t. but it would be convenient if we could summarize the current cost of borrowing in a single number. rt . which can be seen in this example around the 4 year and 8 year marks. giving an inverted curve (ﬁgure 5. which is the leftmost point of the yield curve at time t. but if current rates are high and expected to fall.2. the rate we get is the yield R(t. Figure 5. where ∆t is a small time increment.5 shows an example short rate over ten years. free of any other parameters. t). t + ∆t) . The instantaneous rate is not only an important process in the interest rate market. with all the other bonds extrapolated from it.1. t + ∆t):
R(t.5: Instantaneous rate
We can sometimes see an interaction between the short rate and the bond prices if they are correlated. when the short rate gets high and the bond price dips.
Instantaneous rate
But what is the price of money now? The yield curve gives us an idea of the rate of borrowing for each term length. If at time t we borrow over the period from t to t + ∆t. the high short rate at t = 4 even exceeds the increased yields on longer bonds. A good model should be able to cope with both these possibilities. What we can do is look at the current rate for instantaneous borrowing. ∂ rt = − log P (t.
.4). We call this value the instantaneous rate. THE INTEREST RATE MARKET
103
rates. the yield curve can become ‘inverted’ and long bond yields will be less than short bonds (ﬁgure 5. ∂T
The instantaneous rate is just a process in time. t + ∆t) = − log P (t. but many models are based exclusively on its behavior. corresponding to the evolution of the 10year bond in ﬁgure 5. t). bond prices might be lower when the short rate is higher.5. or short rate.4).
say k units. and will give us a payment of one dollar at time T2 . that is agreeing. at time t. rt . called simply the forward rate. T1 ) at time t. T ) for our particular example is shown in ﬁgure 5.6. that is
f (t. The translation f (t. of the T1 bond. But the formula for R(t. to make a payment at a later date T1 and receive a payment in return at an even later date T2 . T ). T ) + (T − t) ∂R (t. then as the increment ∆t became smaller this would converge to a forward rate for instantaneous borrowing. but the other points of the two curves will generally be different. or face arbitrage. P (t. ∂T
This tells us that the forward rate curve is higher than the yield curve. T ) we can recover the prices P (t. INTEREST RATES
Forwards
The shortrate process. What we require is a natural extension of rt which brings back the onetoone mapping to the prices P (t. and will require us to make a payment of k at time T1 . T ) can be differentiated and rearranged to show that
f (t. T ). T ) and the yields R(t. T ). To give the contract nil initial value. T2 ) − kP (t. The corresponding (forward) yield is then
− log P (t. The translation also entails a loss of information. is not a onetoone mapping from the discount price curve P (t. yet still preserves the idea of instantaneity. T ) = − ∂ log P (t. ∂T
This rate. T ). T ). buying a T2 bond and selling a quantity. if the yield curve is increasing. the instantaneous rate.3). But what forward price should we pay? There is a way of replicating this contract by. at time T = t. Superﬁcially. T ). T1 ) . T2 ) . T2 ) − log P (t. This deal has initial cost P (t.
f (t. t) = rt . given the forward rates f (t. at time t. Yet the instantaneous rate is convenient to work with.
. T ) = R(t. and lower than it if the yield curve is inverted. Consider forward contracts. Just giving rt with no extra prescription on how bond prices can move will not in general be enough to recover P (t. As we might expect. the forward price of purchasing the T2 bond at time T1 . T1 )
This k must be. the ‘forward’ rate for borrowing now. T2 − T1
Were we to choose T1 and T2 very close together. we must set k to be
k= P (t. say T1 = T and T2 = T + ∆t. is the forward price of instantaneous borrowing at time T . We are really just striking a forward on the T2 bond. Indeed the yield curve and the forward curve agree at their leftmost point. T ) and the yields R(t. the forward rate curve resembles the bond yield curve (ﬁgure 5.104
CHAPTER 5. is exactly the current instantaneous rate.
But unlike rt .
the bond prices can be given in terms of the forward rates or the yields:
T
P (t. The Itˆ manipulao
. T ) and f (t. The average yield of the bond over its remaining lifetime is R(t. T ). The price at time t of the T bond which pays off one dollar at time T is P (t.2. Both of these associated families of rates. t). u)du . ﬁnishing with the value rT at time T . T ) = −
log P (t. T ).6: Forward rate curve at time t = 0
As a function of time rather than maturity the forward rate will not be so smooth.
%
In other words. The secret of this chapter is that we can tackle interest rate models in exactly the same way as stock models. T ) = exp −
t
f (t. for modeling purposes we can choose to specify the behavior of only any one of these three. and the other two will follow automatically. A SIMPLE MODEL
105
Figure 5. T ) . R(t. T −t
And conversely. ∂T
and
R(t. Explicitly
' $
Interestrate market summary The forward rates and the yield can be written in terms of the bond prices as
f (t.2 A simple model
A concrete example is illuminating. T ). T ) = − ∂ log P (t. T ) . The price of instantaneous spot borrowing is rt = R(t.
Summary
We have a market of defaultfree zero coupon discount bonds. but will start with some initial value f (0. T ). and the price now of instantaneous borrowing at time T is the forward rate f (t. contain all the original price information which can be recovered. T ) and evolve as a stochastic process. T ). T ) = exp − (T − t)R(t.5.
5. t) = f (t.
and
&
P (t.
] Unlike the basic stock models. INTEREST RATES
tions are harder but they are not signiﬁcant for the story — just as in BlackScholes. we can write down its integral equation easily enough as
t
rt = σWt + f (0. Itˆ o ∂f tells us it is actually drt = dt f (t. f (t. In BlackScholes. T ) = σWt + f (0. the real work is carried by the martingale representation theorem. as long as we are an Itˆ step away from the price of something. there is a canonical candidate — the cash bond formed by the instantaneous rate rt . but there are now an inﬁnite number of underlying discount bonds. t) + ∂T (t. Simple interest rate model Given an initial T integrable forward rate curve f (0.
Since rt is given by f (t. which itself is independent of the apparent ‘choice’ of instruments to work with. Moreover the forward rates at different maturities are perfectly correlated in their movements as the difference between any two of them. this doesn’t have to pose a problem. but such worries will prove illusory. which
. t)ds. a bounded deterministic function of time and maturity. T ) ‘evaluated’ at T = t. the forward curve evolves as:
dt f (t. the Brownian motion Wt . Bt given by
dBt = rt Bt dt. t) +
α(s. and it is normally distributed.
Tradable securities
Tradables may only be an Itˆ step away. o The forward rate itself is
t
f (t. but with the SDE for the forward rate. T )ds. T ) +
α(s. S). Though chapter six will show the choice of numeraire doesn’t really matter. this rate is not constant but rather is a random process. there were only two canonical tradables (the stock and the bond). T ) − f (t. t). There is only one source of randomness.
for some constant volatility σ and drift α. All the tradables will still turn out to be martingales under the riskneutral measure. not over maturity.
0
Thus the forward rate is normally distributed. T ) = σdWt + α(t. is purely deterministic. B0 = 1.
0
[Technical trap: the SDE for rt is not just the SDE for f (t. However as chapter four has shown. t). T ). That is. We have set the market up not with an SDE for the price of any asset. To pick just two of these tradables to work with would seem to favor that pair over the rest. but what are they? One is obvious — we o want a numeraire. T )dt.106
CHAPTER 5. and that is a process over time.
and ﬁnding Zt doesn’t pose any real problem. u)du +
0 0 s
α(s. T ): t T t T
Zt = exp − σ(T − t)Wt + σ
0
Ws ds +
f (0. Bt = exp 0t rs ds . T ) given by P (t.
Replicating strategies
Suppose then we wanted to replicate some claim X at a time horizon S less than T (so that the T maturity bond doesn’t vanish on us). such that dEt = φt dZt . Fixing T .
(3) Find a previsible process φt . this (random) Bt and the T bond price P (t. What about another tradable? Here.
& %
−1 First we tackle the complicatedlooking discounted bond price process Zt = Bt P (t. so at least a ﬁrst guess would be to follow it here as well:
' $
Three steps to replication (interestrate market) (1) Find a measure Q under which the T bond discounted by the cash bond Zt = −1 Bt P (t.2. u)duds . Itˆ gives us the SDE for Zt as o
T
dZt = Zt
−σ(T − t)dWt −
t
1 α(t.
This will be our tradable numeraire. T ) = exp − tT f (t. u)duds . A SIMPLE MODEL
107
admits the possibility of it being negative sometimes. T ) = exp − σ(T − t)Wt +
f (0. there’s an embarrassment of tradables. u)du or
T t T
P (t.5. as mentioned earlier. In chapters two. Later we will show models which overcome this. but for the moment we’ll pay this price for simplicity.
−1 (2) Form the process Et = EQ BS XFt . Now for the cash bond. Wt .
. u)duds .
Noting that the σ(T − t)Wt term’s differential can be handled with the product rule and everything else inside the exponential is either constant or easy to differentiate. T ) is a martingale. u)du +
t 0 t
α(s. T ) are both adapted to the same Brownian motion. 2
Now we are on familiar ground. we have the price of the T maturity bond P (t. which has the slightly daunting expression
t t t t
Bt = exp σ
0
Ws ds +
f (0. but let’s pick one. Though we are used to the cash bond Bt being deterministic. u)du +
0 0 s
α(s. three and four we had a threestage replicating strategy. u)du dt + σ 2 (T − t)2 dt .
The strategy has an initial cost of V0 = EQ (BS X) and has a terminal value VS = X . such that Wt = t Wt + 0 γs ds is a QBrownian motion. T ) + ψt Bt = Bt Et .
The process Zt has no drift. and because σ(T − t) is bounded up to time S .108
CHAPTER 5. namely −1 Et = EQ (BS XFt )
But since Et is a Qmartingale just as Zt is. equivalent to P. u)du. we take:
Step three
Using the martingale representation theorem to link them via an F previsible process φt :
Et =
−1 EQ (BS X) + t 0
φs dZs . Zt is a Qmartingale.
What is our trading strategy? At time t. T ) + ψt dBt and thus this portfolio is self −1 ﬁnancing. which exactly hedges the claim.
• hold φt units of the T bond P (t. the technical conditions of CMG are satisﬁed ˜ — our candidate passes and we have a measure Q. it is also true that dVt = φt dt P (t. The candidate is clearly
1 1 γt = − σ(T − t) + 2 σ(T − t)
T
α(t. Arbitrage has won through.
t
And since it is bounded up to time S . T ) • hold ψt = Et − φt Zt units of the cash bond Bt . Option price formula (interest rate) The price of X at time t is
−1 Vt = Bt EQ (BS XFt ).
The undiscounted value of this portfolio at time t is
Vt = φt P (t. INTEREST RATES
Step one
We have an SDE for Zt and we want to see if we can ﬁnd a change of measure drift γt for the Brownian motion which makes Zt driftless.
Step two
−1 This gives us Et as the conditional Qexpectation of the discounted claim BS X . The SDE for the discounted price Zt now becomes
˜ dZt = −σZt (T − t)dWt .
As before.
.
as expected. o
S
dYt = Yt
−σ(S − t)dWt −
t
1 α(t.
t
S then we have dYt = −σYt (S − t)(dWt + γt dt).
This discounted process must be a Qmartingale — it’s tradable and.
. One direct from its SDE . And the other indirect. What condition? −1 Consider the discounted process of the S bond. S) we would have an arbitrage engine capable of locking in unlimited proﬁts. T ) behaved like any of the tradables of chapter four. for example. S) = 1 as a claim to be hedged via the cash bond and the T bond. If the hedge price were ever. P (t. There is no obvious reason why they should be the same given our original model. and found a change of measure particular to that. S) = EQ (BS Ft ). We don’t want free lunches. Multiplying the formula for γt by σ(T − t) and differentiating with respect to T . Yet all claims which paid off at time S before T could be hedged. the T maturity bond. S). S) agree. That is we should impose on our real world model some suitable condition to make the various ways of getting at the price P (t. so we should assume that the real world forbids them. So we have two ways of pricing the S bond at time t. even those. A SIMPLE MODEL
109
No free lunches
So far. We picked a particular bond. So the drift term of the S SDE above must be zero: γt = γt . we have just another stocko type model. The chosen pair Bt and P (t. Reworking the Itˆ from before we have. viewing X = P (S. And yet the same they must be. Yt = Bt P (t. or in terms of the QBrownian motion we had before: S ˜ dYt = −σYt (S − t) dWt + (γt − γt )dt . u)du dt + σ 2 (S − t)2 dt . So γt must be independent of T . Yt = Bt P (t. or in other words ∂T = 0. 2
S If we deﬁne γt to be
1 1 S γt = − σ(S − t) + 2 σ(S − t)
S
α(t. we get: Restriction on the drift In an arbitragefree market. from the risk−1 −1 free hedging construction. say. T ) = σ 2 (T − t) + σγt .5. which are identical to bond of other maturities. u)du. so good — even though the Itˆ work was harder. But something should worry us. S). the drift α(t. less than P (t. T ) satisﬁes
α(t. Here is the restriction we require — our arbitrary choice of T must not have af∂γ fected the process γt .2.
4). replication provides the price. but now it is expressed as the sum of a particular function (σ 2 (T − t)) and a process which has no maturity dependence at all (σγt ). T ) − σ(T − t)dWt + rt − σ(T − t)γt dt . Our second point is that there is a price to pay for this success. We know that every security in the market has to have the same market price of risk. There was only one Brownian motion and that is what matters. We can think of the SDE for P (t. T ) has vanished. γt stands revealed as the market price of risk (see section 4.110
CHAPTER 5. at this level. It says that there are restrictions on the drifts which the forward rates can have if there is to be no arbitrage. T ) in terms of ˜ the QBrownian motion. If we write the original SDE for f (t. that is. Firstly there is a measure Q which makes a martingale not just out of one discounted bond.
Written this way. T ) may have started off as a general deterministic function of both time and maturity. the drift α(t. But α(t. So we weren’t free to pick α(t. in the stock market. T ) = P (t. T ) = σdWt + σ 2 (T − t)dt. but each and every discounted bond simultaneously. All claims can be coherently hedged by the underlying bonds. Once more. unlike BlackScholes. But even if our success has brought slight complications. T ) under P as
dt P (t. We worried about the embarrassment of bonds to choose. we have:
˜ dt f (t. And if one bond can be brought into line by a change of measure then so can they all. we have nonetheless succeeded. they are roughly in step in the ﬁrst place. T ) must be recoverable by a change of measure γt which has no dependence on T . this is actually familiar ground. T ) as any function of t and T — we must. the values of the bonds P (t. which explains why γt does not depend on the maturity T chosen. If. If we freeze time and look at just one t.
As we expect from a BlackScholes upbringing. In another sense. The drift α(t. have some structure to the original real world drift. Two things stand out.
. but we needn’t have. INTEREST RATES
This equation is saying something we did not encounter. Wt . Most general functions cannot be written in this way. T ) are just deterministic transformations of each other. We have a model with stochastic interest rates which is still arbitragecomplete.
0
0 ≤ t ≤ T. the forward rate for each maturity T evolves as
t t
f (t. T ). T ) = exp − σ(T − t)Wt +
t T t
f (0. T ) = f (0. The ﬁrst two conditions make sure that the forward rates f (t. the prices P (t. T )dWt + α(t. The general HJM model posits very few overarching conditions on the σ and α. T ) can depend on the history of the Brownian motion Wt and on the rates themselves up to time t. T )ds. but here only a single process.3. t) + 1 σ 2 t2 . the yields R(t. these technical conditions are shown in the box. T ) = σ(t. T ) = σ Wt + f (0.
The volatilities σ(t.5 we will allow the decoupling that comes when rates can move with less than perfect correlation. T ) are well deﬁned
. HeathJarrowMorton (HJM) is a powerful. T ) + σ 2 T − 1 t t. For any ﬁxed maturity T . u)du + 1 σ 2 t3 . In section 5. rt = σ W
2
&
%
5. T ) and the drifts α(t. T ) and the forward rates f (t. 2
Bt = exp σ
0
˜ Ws ds +
t 0
f (0. T ) are equivalent. T )dt. We know the basic idea — all three descriptions of the yield curve. The incremental changes of all forward rates. T ) and its own drift α(t. the forward rate evolves according to its own volatility σ(t.
or in differential form
dt f (t. T )dWs +
α(s. so we select one and specify its behavior. will drive each and every rate. Singlefactor HJM model Given an initial forward rate curve f (0. forward and short rates are given by:
˜ P (t. SINGLEFACTOR HJM
'
111 $
˜ Bonds and rates in terms of the QBrownian motion Wt
The bond prices. technically rigorous interestrate model based on the instantaneous forward rates f (t. a PBrownian motion Wt . Collected. T ) +
0
σ(s. T ). but imposes piecemeal technical constraints from time to time. Our formal description is vague about the precise properties of the volatility and drift functions.3 Singlefactor HJM
From the particular to the general. and thus all yields and all bond prices are perfectly correlated. 2 ˜ t + f (0. and simpliﬁed somewhat. 6
˜ f (t.5. T ). T ). u)du + 1 σ 2 T (T − t)t .
u)duds . T 0 u 0 σ(t. the bond B is a stochastic process satisfying the SDE
t
dBt = rt Bt dt. u)du +
0 0 s
α(s. t)dWs +
α(s. formed by starting with one dollar at time zero and reinvesting continually at this rate.
B0 = 1.
or
Bt = exp
0
rs ds . the bond prices themselves are
. T ) are previsible and depend only on the history of the Brownian motion up to time t. T ). the choice of numeraire is arbitrary — but algebraic convenience certainly points to a canonical choice. t) (which need not be Markov). u)dWs du can be interchanged to 0 s σ(s.
Integration then gives us
t t s t t t
Bt = exp
0
σ(s. f (0. T ). t) +
0
σ(s. T ) are a onetoone transformation. Since the forward rates f (t. Singlefactor HJM: conditions on the volatility and drift We assume that
• for each T . the processes σ(t. The last two conditions will be used for a Fubinitype result that the stochastic differential of the integral of f (t. Given these box conditions. is deterministic and satisﬁes the condition T that 0 f (0. u)dWt
• the volatility σ has ﬁnite expectation E
du.
0
The simplest cash product is then the account. INTEREST RATES
by their SDE. • the initial forward curve. u)du dWs .
Numeraire
As chapter six will show. u)du dWs +
f (0. u)dtdu. t)ds. namely:
t t
rt = f (0. T )dt and 0 α(t. In other words. T ) with respect to T is the integral of the stochastic differentials of f . T ) and α(t. T )dt are ﬁnite.112
CHAPTER 5. and are good integrators T T in the sense that 0 σ 2 (t.
Here we used the last technical condition of the HJM box to say that the integrals t t t t of 0 s σ(s. • the drift α has ﬁnite integral
T 0 u 0 α(t. Our description of the forward rates f (t.
Bond prices
We need tradable assets — and we have them. the ﬁrst three conditions of the HJM model (C1–C3 in their paper) are satisﬁed. u)du < ∞. T ) allows us to write down an integral equation for the instantaneous rate rt = f (t. We have a numeraire. the bonds P (t. or bond.
u)du +
t 0 t
α(s. T ) Σ(t.
t
revealing the variable Σ(t. u)du ) and time T (namely one).
which will be continuous in t and T . T ) = Bt P (t. T ) = Z(t. −1 our attention focuses on the discounted asset price — that is. u)du dWs +
f (0. T )dt < ∞ holds (see section 3. it is sufﬁcient that the expoT nential martingale condition EQ exp 1 0 Σ2 (t. As everywhere else. Itˆ handleturning then o
T
1 2 2 Σ (t. T ) equal to
t T t T t T
exp −
0
σ(s. T )dWs −
f (0.
where Σ(t. T ) then we have the bond price P (t. u)duds .3. u)du dt . Z(t. although awkward. u)du .
Change of measure
In the usual way. has the right values at time zero T (namely exp − 0 f (0. T )Σ(t. T ) = exp −
t
f (t. u)duds . such that Wt = t Wt + 0 γs ds is QBrownian motion. By combining the above expressions for the cash bond and the bond price itself.
Reassuringly this expression. T ) − 2 Σ(t. For this to be a proper Qmartingale. u)du. T ) = Z(t. we get
t T t T
Z(t. If we integrate the original equation for the forward rates f (t. T ) is just notation for the integral − gives the SDE —
dt Z(t.
which is driftless. 2
. T )dWt +
σ(t. T ) −
α(t.
t T
1 2 We need the technical CameronMartinGirsanov theorem condition that EP exp 2 0 γt dt ˜ is ﬁnite. T ) = exp
0
Σ(s. The change of measure drift (market price of risk) is
1 1 γt = Σ(t. T ) to be the logvolatility of P (t. T )
T
α(t. T ). u)du −
0 T t 0 s
α(s. Then there will be a new measure Q equivalent to P. T ). u)du.5). we want to make the discounted bond into a martingale by changing measure. T )dWt .
Discounted bonds
Let’s ﬁx one particular maturity T to work with for the moment. SINGLEFACTOR HJM
113
contained in the forward rate information as
T
P (t.5. The SDE of the discounted bond under Q is then
˜ dt Z(t.
we choose to use a bond with maturity T larger than S . • hold ψt := Et − φt Z(t. But we ought to check that we can produce replicating strategies for claims. but only on the volatility Σ (itself a function of σ ). for simplicity.
Replicating strategies
We’ve jumped slightly ahead of ourselves. That is. T ) is never zero before T . Our trading strategy will be a combination of both a holding in the T bond and a holding in the cash bond Bt .) Suppose.
The value of this portfolio at time t is
−1 Vt = Bt Et = Bt EQ BS XFt .
For the martingale representation theorem to be used. in which case. T ) = P (t. rather than the raw claim X . we have found the martingale measure and the process for P (t. This gives us that
t
Et = E0 +
0
φs dZ(s. we apply the representation theorem to the martingale Z(t. The behavior of the price P under the martingale measure does not depend on the drift α.
The concrete model of section 5. T )dWt + rt dt .2 partially spoilt the surprise.
for some F previsible process φ. we deﬁne Et to be the Qmartingale
−1 Et = EQ BS XFt . and roll over shortterm bonds from section to section. our second step to replication is to form the conditional Qexpectation −1 of the discounted claim BS X . INTEREST RATES
Bond price SDE Under this martingale measure. even here with a general interestrate model such as HJM. (Unless we split the timeperiod up into shorter subsections. Just as the BlackScholes stock model under Q had no dependence on the original stock drift µ. T ) under it. If we are going to hedge this with a discount bond maturing at date T . we also need that the bond volatility Σ(t. Speciﬁcally.114
CHAPTER 5. T ) Σ(t. the bond price P now has the stochastic differential
˜ dt P (t. we
• hold φt units of the T bond at time t. but we have our BlackScholes like result. our only restriction is that S should come before T — we cannot hedge a longterm product with a shorterterm instrument. As before.
. T ). T ) and the discounted claim process Et . Suppose we have a claim X which pays off at time S . T ) units of the cash bond at time t.
Which is ensured by the representation of Et in terms of φt .
Differentiating with respect to T .
Just as we saw in the simple example (section 5. T ) = σ and Σ(t. Z(t. So we have a restriction on the bonds’ Pdrifts.
T t
α(t. or else there will be an arbitrage opportunity. T ) = −σ(T − t). the real world drift α(t. T ) − Σ(t. All bond prices are just the expectation under Q of the instantaneous rate discount from t to their maturity. for all maturities T .2. T ).
The martingale measure brings a pleasant simplicity. T )γt . SINGLEFACTOR HJM
115
The strategy (φt . the forward rate and the instantaneous rate are then. Under this riskneutral measure. T ) γt − Σ(t. S) = Bt P (t.
t ≤ S < T.
S
P (t. −1 What about the discounted S bond. all the other (discounted) bonds are now martingales under the same Q. u)du = 1 Σ2 (t. T ) + σ(t.5. T ) = −σ(t. T )Σ(t. Thus its worth at time −1 t must be Bt EQ (BS Ft ). that is
α(t. where σ(t. S) = EQ BS Ft . then its value at time t is
Vt =
&
−1 Bt EQ BT XFt T
= EQ
exp −
t
rs ds X Ft
.
%
Arbitragefree market
But the S bond is simply a claim of X = 1 maturing at time S .
Exactly as in section 5. it must be the case that. S)? This can now be written as
−1 Z(t.
.7) if
dEt = φt dt Z(t. the market price of risk has to be the same for all bonds. T )γt . T ) . T ) = σ(t. ψt ) is selfﬁnancing. S) = EQ
exp −
r
ru du
Ft
. T ) cannot be too different from the riskneutral value of −σ(t. ψt ) will be selfﬁnancing if dVt = φt dt P (t. T )Σ(t.2). T ). In particular. T ) + ψt dBt . The portfolio (φt . Or more fully. or equivalently (as in section 3. we see that α(t. Thus:
' $
Derivative pricing If X is the payoff of a derivative maturing at time T . In other words. 2
t ≤ T. Which means that their drifts under P are restricted by the need to be a simple change of measure away from a martingale.3.
The ﬁrst condition is actually necessary and sufﬁcient for there to be an equivalent measure under which every single discounted bond price is a martingale.
for all t ≤ T . T ) modiﬁed only by the ‘onedimensional’ displacement γt σ(t. which means that all risks can be hedged using the martingale representation theorem. t) +
0
˜ σ(s.
1 • the expectation E exp 2 T 0 2 γt dt is ﬁnite. the drift is forced to take the value −σ(t.
t
rt = f (0. T )dWt − σ(t.
Model conditions
We have been accumulating technical conditions as we have swept through.
0
Like the bond price itself. which guarantees the absence of arbitrage. T ) . t < T . T ) and Σ(t.116
CHAPTER 5.
2
The importance of the ﬁrst condition in this box is the constraint it places on the drift α(t. but are solely expressed in terms of the volatilities σ and Σ. t)Σ(s. T ) = σ(t. T ) = σ(t. t)ds. Unlike simple asset models. T ) γt − Σ(t. The second condition is equivalent to asserting that the change of measure is unique. T ) dt is ﬁnite. As the process γt is only a function of time and not of maturity. T 0
1 • and the expectation exp 2
γt − Σ(t. Given that σ(t. not all drift functions α(t. t)dWs −
t
σ(s. T )Σ(t. T ) are determined by the forward rate volatilities. T ). such that α(t. They are summarized in the box below.
. for every maturity T . T ). these expressions no longer depend on the drift at all. ω). T )Σ(t.
• the process At = Σ(t. Singlefactor HJM: market completeness conditions It is required that
• there exists an F previsible process γt . T )dt. INTEREST RATES
Forward and short rates under Q
˜ dt f (t. T ) is nonzero for almost all (t. T ) are allowable. The last two conditions are technical requirements for CMG to operate and to make sure that Z is a martingale under the new measure. the only degree of freedom for the drift comes from the oneparameter γt process.
To prove this requires choosing the forward volatility surface σ(t.4 Shortrate models
Shortrate models are popular in the market. They have evolved from various historical starting points — some emerging from discrete frameworks. This is possible for any general short rate rt . u)du = − log P (t.4.
There is a theorem:
. as in the HJM model. T ) = EQ
exp −
t
rs ds X Ft
.
and the value at time t of a claim X maturing at date T is
T
V (t.
HJM in terms of the short rate
It is not immediately clear that this is an HJM model. t) are deterministic functions of space and time. T ) where g(x. t). which is why we used this framework in the ﬁrst place. t) and drift ν(rt . T ) so that the resulting short rate from the HJM model is exactly the same as the original process rt . t. Demonstrating that is the purpose of this section. which typically are Markovian. All however are HJM models.3. t. t)dWt + ν(rt .
The paradigm of shortrate modeling is to work within a parameterized family of processes. Rolling up the periods gives rise to a cash bond process Bt = exp 0 rs ds . others from equilibrium models — and are often presented in a simple hierarchy with no apparent connection to any overarching model. T ) = EQ
exp −
t
rs ds
Ft
.5. T ) = g(rt . T ) is the deterministic function
T
g(x. The parameters are chosen to best ﬁt the market. A shortrate model posits a riskneutral measure Q and a shortrate process rt . but it’s easiest to show in the special case where rt is a Markov process. As with the equations at the end of section 5. and then the above expression for Vt is calculated to price the claim X . t) and ν(x.
where ρ(x. The model is that instantaneous borrowing can take place at rate rt for an inﬁnitesimal t period. t. Suppose that that rt is a Markov diffusion (though not necessarily timehomogeneous) with volatility ρ(rt . SHORTRATE MODELS
117
5. In particular. the bond prices are given by
T
P (t. they are often used to price derivatives which depend only on one underlying bond. Then tT f (t. T ) = − log EQ
exp −
t
rs ds
rt = x . t)dt. That is
drt = ρ(rt . And there is a mathematical transformation that makes these two alternative descriptions equivalent.
T ) = ρ(rt .
0
This formula is not necessarily simple. t. T ). 0. which gives us the result. given as
t t
rt = f (0. t) +
0
σ(s.
Ho and Lee
Now for the accepted hierarchy of shortrate models: starting with Ho and Lee. the martingale measure. ∂x∂T ∂g Σ(t. T ) and Σ(t. ∂T
This volatility structure and initial curve then identiﬁes an HJM model for this market with the same short rate under Q. t. T ). t)dt + dt + ρ (rt . T ) = ∂x∂T ∂t∂T 2 ∂ 2 x∂T
The volatility term must match σ(t. with rt in terms of the HJM volatilities σ(t. INTEREST RATES
Shortrate model in HJM terms The required volatility structure is
∂ 2g (rt .
Short rate in terms of HJM
Conversely.3).
for some θt deterministic and bounded. T ) is given by
f (0. ∂x σ(t. by thinking of the forward rate f (t. t) (rt . T ) = ∂g (r0 . we ﬁnd via Itˆ o
1 g(x. t)Σ(s. T ) as and using Itˆ to deduce that o
∂ 2g 1 ∂ 3g 2 ∂ 2g ρ(rt . T ) = −ρ(rt . dt f (t. In addition. The equation for the bond price in terms of rt holds (see near the end of section 5.
.
and
We can see why this is so. and σ constant. the initial forward rate curve f (0. t. t)dt. T ). it is also true that HJM models are shortrate models. The question we should immediately ask is. which HJM model corresponds to this? Following the mechanics from earlier. In its shortrate form. t)ds. as Ho and Lee model The short rate is driven by the SDE:
drt = σdWt + θt dt. T ) = x(T − t) − σ 2 (T − t)3 + 6
T t
(T − s)θs ds. t)dWt + ν(rt . T ). t)
∂g ∂T (rt . T ). Ho and Lee gives the SDE for rt under Q.118
CHAPTER 5. T ). t. t)dWs −
σ(s.
5.4. SHORTRATE MODELS
2
119
∂ g Thus σ(t, T ), the HJM volatility surface, is simply σ ∂x∂T = σ . Thus the volatility surface is constant, depending on neither time nor maturity. We can fully specie the HJM model under Q as
Ho and Lee model in HJM terms
dt f (t, T ) = σdWt + σ 2 (T − t)dt ∂g 1 (r0 , 0, T ) = r0 − σ 2 T 2 + ∂T 2
T 0
with
f (0, T ) =
θs ds.
Equivalently, we can provide the evolution of the bond prices P (t, T ) under Q:
T
P (t, T ) = exp − σ(T − t)Wt +
t
1 f (0, u)du + σ 2 T (T − t)t . 2
This model is the (general) singlefactor model with constant volatility, and is actually the simple model of section 5.2. If used in the shortrate form, then σ sets the volatility of all forward rates and θs allows matching to any initial forward curve via T the identity f (0, T ) = r0 − 1 σ 2 T 2 + 0 θs ds. 2 It is a simple model, and its simplicity tells against it — the forward rates and the short rate rt can go negative occasionally, and go to inﬁnity in the long term. And not just of course under Q, but under any equivalent measure P as well. Many other models expend much effort just to avoid these pitfalls. But it is not that simple a model — the HJM formulation allows a description of how the real forward curve can move over time. Given any previsible process γt , the forward rates can move as
dt f (t, T ) = σdWt + σ 2 (T − t) + σγt dt,
with
drt = σdWt + (θt + σγt )dt.
So short rates can have a wide range of possible drifts under P the real world measure, not just the simple deterministic drift θt . The restriction with Ho and Lee lies not there but in the implication that σ(t, T ) = σ . Two extra things need mentioning. First, the bond price and the cash bond price are both lognormally distributed and thus the BlackScholes formula can still hold (as hinted at in section 4.1, and shown in section 6.2). And second, there is a straightforward generalization to a deterministic shortrate volatility
drt = σt dWt + θt dt,
with a corresponding HJM formulation
2 dt f (t, T ) = σt dWt + σt (T − t)dt,
with the initial forward rate curve given by
T
f (0, T ) = r0 −
0
2 σs (T − s)ds +
T 0
θs ds.
120
CHAPTER 5. INTEREST RATES
The extra freedom here is to allow the volatility surface to depend on time, but not on maturity. For that we require something else...
Vasicek/HullWhite
Next in the accepted hierarchy is to allow the short rate’s drift to depend on its current value. Vasicek model We model the short rate (under Q) as:
drt = σdWt + (θ − αrt )dt
for some constant α, θ and σ . The SDE is composed of a Brownian part and a restoring drift which pushes it upwards when the process is below θ/α and downwards when it is above. The magnitude of the drift is also proportional to the distance away from this mean. Such a process is called an OrnsteinUhlenbeck or OU process. We can use Itˆ ’s formula to check that the solution to this, starting r at r0 , is o
rt = θ/α + e−αt (r0 − θ/α) + σe−αt
t 0
eαs dWs .
¯ As it happens, rt can be rewritten in terms of a different QBrownian motion W as rt = θ/α + e
−αt
(r0 − θ/α) + σe
−αt
¯ W
e2αt − 1 2α
,
so that rt has a normal marginal distribution with mean θ/α + exp(−αt)(r0 − θ/α) and variance σ 2 (1 − exp(−2αt))/2α. As t gets large, this converges to an equilibrium normal distribution of mean θ/α and variance σ 2 /2α. This does not mean that the process rt converges — it doesn’t — only that its distribution converges.
Figure 5.7: An OU process, with σ = 2θ = 2α = 1
What HJM model are we in? Again we can use Itˆ to ﬁnd g(x, t, T ), and thus o σ(t, T ) and f (0, T ). In this case
5.4. SHORTRATE MODELS
121
Vasicek model in HJM terms
σ(t, T ) = σ exp − α(T − t) , σ2 (1 − e−αT )2 . 2α2
with
f (0, T ) = θ/α + e−αT (r0 − θ/α) −
Now we can see an advantage over Ho and Lee — where Ho and Lee failed to introduce a maturity dependence into the volatility surface, this model can. Thus this model is capable of calibration to a richer set of observed volatilities. Note how the volatility σ(t, T ) is derived from both the drift and volatility of the short rate under Q. In order to describe an HJM model, we need two degrees of freedom for the volatility — one for time and one for maturity. The short rate description doesn’t abandon the second degree of freedom; it encodes it in the relationship between its volatility ρ(rt , t) and its drift ν(rt , t). The drift of rt under Q is a vital part of the description. But only under Q. The Vasicek model, unlike Ho and Lee, may be mean reverting under Q, but both models are in fact capable of mean reversion under P. Some care has to be taken — the introduction of the extra parameter α does give Vasicek a richer set of allowable Pdrifts than Ho and Lee, but this richness involves maturity. Simple timedependent considerations will not in general prejudice one over the other. Because it is possible to ﬁnd a change of measure γt which gives mean reverting behavior to Ho and Lee, Vasicek is not the inevitable choice if in the real world mean reversion is observed. In practice it will be the volatility of the entire curve, rather than the drift of the short rate that forces one over the other. As before there is a natural generalization to
drt = σt dWt + (θt − αt rt )dt
where σt , θt , and αt are deterministic functions of time. As rt is still a Gaussian process with normal marginals, so f (t, T ) is Gaussian and the bond prices have lognormal marginals. In this case, the HJM volatility and initial forward curve are
T
σ(t, T ) = σt β(t, T ), f (0, T ) = r0 β(0, T ) +
where
T 0
β(t, T ) = exp −
t T 2 σs β(s, T )
αs ds ,
T
and
θs β(s, T )ds −
β(s, u)du ds.
s
0
The normality of the forward rates f (t, T ) is both good news and bad news. In its favor, it means that the bond prices P (t, T ) are lognormally distributed, so that the lognormal option pricing results of section 6.2 all hold. On the other hand, both the instantaneous rate and the forward rates can go negative from time to time. Depending on the parameters, this can happen more or less rarely — the next model rectiﬁes this defect.
INTEREST RATES
CoxIngersollRoss
The model is a meanreverting process. then the process actually stays strictly positive.) Then the function g(x. T ) to be the solution of the Riccati differential equation
∂B 2 = 1 σt B 2 (t. which is (− log P (t. so allowing the drift θt to dominate and to stop rt from going below zero.8: Autoregressive σ = 1. CoxIngersollRoss model The instantaneous rate’s SDE. is
√ drt = σt rt dWt + (θt − αt rt )dt.
where σt .122
CHAPTER 5.
Figure 5. 2
This process is called autoregressive. this equation has no analytic solution. which pushes away from zero to keep it positive (see box). under Q. α = 2
It is difﬁcult to ﬁnd an explicit pathwise solution for rt . T ) − 1. but it has been well studied numerically. T ). The volatility term is set up to get smaller as rt approaches zero. T ) = 0.
then the volatility structure can be expressed as
. 2 ∂t B(T. t. but we can solve a useful partial differential equation (PDE) . T ) +
t
θs B(s. The drift term is a restoring force which always points towards the current mean value of θt /αt . T )ds. t. T )rt = x) can be written in terms of this solution B(t. T ). θ = 2. and αt are deterministic functions of time. T ) as
T
g(x.
(In general. T ) be
∂B ∂T (t. Firstly deﬁne B(t.
2 As long as θ satisﬁes θt ≥ 1 σt . T ) + αt B(t. T ) = xB(t. θt .
Letting D(t.
there is some volatility surface σ(t. √ Σ(t. . .
BlackKarasinski
Another way round the problem of keeping the short rate positive is to take exponentials. and additionally is always positive. Correspondingly each T bond forward rate process has a volatility σi (t.
5. This model is still HJM consistent. T ). the bond price P (t. T ) = −σt rt B(t. Explicitly: BlackKarasinski model The process Xt is
dXt = σt dWt + (θt − αt Xt )dt. Wn (t). This model is an extension of the BlackDermanToy model and starts by taking Xt to be the general OU process of the Vasicek model. . and to have strong correlations with some bonds and weaker correla
.5. T ). A multifactor model involves driving the various processes by a collection of independent Brownian motions. T ) = σt rt D(t. We also know that T Xt is normally distributed. we will have n Brownian motions to work with: W1 (t). . T ) +
0
θs D(s. The rate itself also drifts towards a mean.
So the logarithm of the rate drifts towards the current mean of θt /αt .5. t. This allows different bonds to depend on external ‘shocks’ in different ways. that is. and αt are deterministic functions of time. The instantaneous rate rt is then assumed to be
rt = exp(Xt ). t rs ds is awkward to examine analytically. T ). More details of such models are in section 6. In an nfactor model. that assumption is too coarse.
As usual.3. For many applications. T )ds. MULTIFACTOR HJM
123
CoxIngersollRoss in HJM terms and with
√ σ(t. T ) = exp −g(rt . T ) for each Brownian factor Wi (t). However. T ) which generates a singlefactor HJM model which has the same instantaneous rate as given above. especially if we are trying to price something which depends on the difference of two points on the yield curve. T ) = r0 D(0.5 Multifactor HJM
The drawback of the singlefactor model is that all the increments in the bond prices are perfectly correlated. θt .
where σt . so that rt is lognormal.
T
f (0. T ) has the form P (t.
That is. T )ds. n is 1. t)ds. The instantaneous rate rt = f (t. u)du. where Σi (t. T ) +
i=1
σi (s.
In the singlefactor model. but with one difference. t)dWi (s) +
α(s.
which is to say that the forward process starts with initial value f (0. T ) and f (t.
This is just as in the singlefactor case. is deterministic and satisﬁes the condition T that 0 f (0. u)dWi (t)
• each volatility σi has ﬁnite expectation E
du. T 0 u 0 σi (t. as a
. the total instantaneous square volatility of f (t. we need a version of the CameronMartinGirsanov theorem for higher dimensions (section 6. u)dtdu. Multifactor HJM: market completeness conditions (1) It is required that
• there exist previsible processes γi (t). T )dt and 0 α(t. • the drift α has ﬁnite integral
T 0 u 0 α(t. t) +
i=1
σi (s. f (0. T ) F previsible and their integrals T 2 T 0 σi (t. T ). S). i=1 n
and
i=1
σi (t.
To make the discounted bond prices into martingales. and the correlation of the changes in the forward rates of the T bond and S bond is exactly one. such that
1 • the expectation E exp 2 T 0 2 γi (t)dt is ﬁnite. T ) and α(t. u)du < ∞.124
CHAPTER 5. INTEREST RATES
tions with others. T )σi (t. The general form of the multifactor HJM model is
n t 0 t
f (t. T ) = f (0. similarly to before as
n t 0 t
rt = f (0. T ). S) are respectively
n 2 σi (t. t) can be written. • the initial forward curve. T ) and is driven by various Brownian terms and a drift. T )dWi (s) +
α(s.3). T ). The drift is now allowed n ‘dimensions of freedom’ away from its riskneutral value. The conditions we need for this to work are shown in the two boxes. From this.
0
0 ≤ t ≤ T. for 1 ≤ i ≤ n. the processes σi (t. T ) is the T integral − t σi (t. and and the covariance of the increments of the two forward rates f (t.
0
The volatility and drift conditions are generalized to: Multifactor HJM: conditions on the volatilities and drift We assume that
• for each T . T )dt are ﬁnite.
2
The modiﬁcation from the singlefactor case here is that the volatility process At which used to be required to be nonzero has been replaced by a volatility matrix process which has to be nonsingular.
Using the multidimensional CMG. ·) is allowed to deviate by any linear combination of the functions αi (t. So Z ’s SDE is (in Qterms)
n
dt Z(t. T ) satisfy
dt Z (t. for every set of maturities T1 < T2 < · · · < Tn . u) Σi (t. • and the expectation E exp 1 2
n T i 0
γi (t) − Σi (t. T ) = Z(t.j=1 is nonsingular for almost all (t.
The SDE for Z now becomes
n
dt Z(t. Multifactor HJM: market completeness conditions (2) We also need that
• the matrix At = (Σi (t. T )
σi (t. The second condition ensures that the resulting driftless discounted bond price is in fact a martingale (a multidimensional equivalent of the collector’s guide to exponential martingales).
where Σi (t. This is still much less than the set of all possible functions.
and every Z(t. ω). . ·).5. T ) dt is ﬁnite. where Wi (t) = Wi (t) + t 0 γi (s)ds. we can ﬁnd a measure Q equivalent to P. u) +
t T t
n i=1
σi (t. T ) =
n
Σi (t. Wn are independent QBrownian motions. Tj ))n i. but it is larger than in the singlefactor case. u) +
i
σi (t. .5. T ) = Z (t. The second condition is the technical requirement of the CMG theorem for γi (t) to be a drift under an equivalent change of measure. T )
i=1
˜ Σi (t. u)du. The discounted bond prices Z(t. t < T1 . . T ) = Z(t. MULTIFACTOR HJM
125
function of T . T )
i=1
Σi (t. T ) dWi (t)
i=1 T
−
t
α (t. . T ) is a Qmartingale in t. As before we ﬁnd that the bond prices themselves have stochastic increments
n
dt P (t. u) du dt . T ) dWi (t) + γi (t)dt . T )dWi (t).
. T )
i=1
Σi (t. under ˜ ˜ ˜ which W1 . α(t. u) du dt . T ) = P (t. u) Σi (t. T ) dWi (t)
T
+ rt −
α (t. T ) is the integral − −1 Bt P (t.
. Mt = M1 (t). INTEREST RATES
Under this martingale measure.
i=1
Derivative pricing and hedging
The actual price of a derivative still has a familiar form: Option price formula (HJM) If X is the payoff of a derivative at time T . T )dWi (t) −
σi (t. Tn all larger than T . as well as the cash bond. then its value at time t is
−1 Vt = Bt EQ BT XFt = EQ T
exp −
t
rs ds X Ft
. . An advantage of the HJM framework is that we are free to choose whichever n instruments we like. Suppose we choose to use bonds with maturities T1 . T )
i=1 n
˜ Σi (t. . If we are going to hedge the claim X with discount bonds. Formally Martingale representation theorem (nfactor) ˜ Let W be ndimensional QBrownian motion. . T )dWi (t) + rt dt . T )dt. if we have an nfactor model. in that dMj (t) = i σij (t)dWi (t). . φn (t) such that 0 ( i σij (t)φj (t))2 dt < ∞. . which has volatility ˜ matrix σij (t) . . As a general rule. and the answer will always be the same. and the martingale N can be written as
n t 0
Nt = N0 +
j=1
φj (s)dMj (s). and suppose that Mt is an ndimensional Qmartingale process. . in order to hedge claims. . . .126
CHAPTER 5. Then if Nt is any onedimensional Qmartingale. T ) =
i=1
˜ σi (t. and the matrix satisﬁes the additional condition that (with probability one) it is always nonsingular. the bond price P and the forward rate f have the stochastic differentials Bond prices and forward rates under Q
n
dt P (t. we need a trading portfolio of n separate instruments. T ) = P (t. . Mn (t) .
Further φ is (essentially) unique. there exists an ndimensional F previsible T process φt = φ1 (t). we must still make sure that all their maturities are later than T . T )Σi (t.
We also need a multidimensional martingale representation theorem.
n
dt f (t.
. T2 .
We cannot hope to describe the hundreds and possibly thousands of traded claims.
This immediately gives a selfﬁnancing strategy φ. . The value of the portfolio at time t is
n
Vt =
j=1
φj (t)P (t. Tj ). and in return to receive a dollar at the later time T2 .
We can now apply the representation theorem in the usual way to the martingale −1 produced from the discounted claim. The part of the martingales Mj (t) will be taken by the discounted bonds Z(t. ψt will be the combination of both an nvector of holdings in the bonds with maturities T1 . contracts which not long ago would have been considered as exotics are now commonplace. Especially in the overthecounter markets. then by the representation theorem. . that is Et = EQ (BT XFt ). and a holding ψt in the cash bond Bt . at the current time t. but we can sketch out the basic types within each area. . −1 In the usual way. INTEREST RATE PRODUCTS
127
A selfﬁnancing strategy φ1 (t). Tn respectively. . . φn (t). the portfolio costs an initial EQ (BT X) and evolves to be worth exactly X by time T .
−1 and its discounted value Et = Bt Vt is n
Et =
j=1
φj (t)Z(t. which is nonsingular by the completeness −1 conditions box. Tj ) units of the cash bond. If we set Et = EQ (BT XFt ). to make a payment of an amount k at a future time T1 . Tj ).
The selfﬁnancing equality for the strategy (as in section 6. . . there is an nvector of previsible processes φt such that
n
Et =
−1 EQ (BT X) + j=1
t 0
φj (s)dZ(s.6 Interest rate products
In recent years. Tj ) i. Tj ).4) is that
n
dEt =
j=1
φj (t)dt Z(t. Tj ) + ψt Bt . Their volatility matrix is given by At = Σi (t. and ψt = Et − j φj (t)Z(t.
5. there has been a great increase in the number of interest rate products available.6.j . We agree. .
Forward contract
This is about the simplest product possible. Tj ) + ψt .5. What should the amount k be?
. We hold φj (t) units of the Tj bond at time t.
n). so its worth at time t is
−1 Vi (t) = Bt EQ (BTi Xi Ft ) = P (t. We can treat each payment Xi separately.
under the martingale measure Q. There are other important cases where this also happens.
• Divide and rule.
For this contract to have null current net worth. T1 ) . T1 )
This price makes sense. the answer is independent of our particular term structure model. T2 ) − kP (t.
where PTi . T2 − T1
For T1 and T2 very close together. as the hedging strategy is static.128
CHAPTER 5.
−1 Recalling that Bt EQ BT Ft is just P (t. Suppose we were. P (t.4). T ). we see that
Vt = P (t. . As long as we bear that in mind. Instead the contract speciﬁes a sequence of payments Xi made at a sequence of times Ti (i = 1.
. but the forward measure. . to buy k units of the T1 bond and sell one unit of the T2 bond. is the Ti forward measure (see section 6. T ) = f (t. T2 ) − log P (t. The initial cost of that deal is zero. INTEREST RATES
According to the pricing formula (under whatever model we are in). . and indeed there are two different ways to keep things clear. T ). T2 ) . T1 ).
Multiple payment contracts
Most interest rate products don’t just make a single payment X at time T . and even on any previous payment. it is just a claim at time Ti . at time t. the value of the claim now is
−1 −1 Vt = Bt EQ BT2 Ft − Bt EQ kBT1 Ft . as saying that the forward yield from T1 to T2 is
− log P (t. we merely choose k at time t to be
k= P (t. and the portfolio pays us k at time T1 (matching the payment we have to make at that time) and exactly absorbs the dollar we receive at time T2 . will have to be changed for each i. On its own. This approach will always work. where Bt is the cash bond
t
Bt = exp
0
rs ds. Each payment Xi may depend on price movements up to its payment time Ti . Ti )EPTi (XFt ). ∂T
The price also gives us a clue to the hedging strategy. In this particular example. this approximates to the instantaneous forward rate of borrowing
− ∂ log P (t. if used. this causes no serious problem. .
where δ is the payment period. Ti ) i=1
Floating rate bonds
A bond might also pay off a coupon which was not ﬁxed.
If we desire the bond to start with its face value. That is. Tn ) + kδ
i=1
P (T0 .
The actual payment made at time Ti is δL(Ti−1 ) = P (Ti−1 . . T )
with worth at time t
−1 Vt = Bt EQ (BT XFt ) = P (t. n) and also pays off a dollar at time Tn . n) of varying amounts.
.
Bonds with coupons
In practice. The price of the coupon bond at time T0 is
n
P (T0 . Ti ) − 1)FT0 . and keep them till the last payment date T . . and also payments at times Ti = T0 + iδ (i = 1. The value to us now. Suppose a bond pays its dollar principal at time Tn . The amount of payment made at time Ti is determined by the LIBOR rate set at time Ti−1
L(Ti−1 ) = 1 δ 1 P (Ti−1 . which is the amount of interest we would receive by buying a dollar’s worth of the Ti bond at time Ti−1 . The amount of the actual coupon payment is kδ . . INTEREST RATE PRODUCTS
129
• Savings account. δ n P (T0 . . Then the payoff is a single payment at time T of
n
X=
i=1
Xi P (Ti .6. T )EPT (XFt ). Ti )−1 − 1. of the Ti payment is
−1 BT0 EQ BTi (P −1 (Ti−1 . . then the coupon rate should be
k= 1 − P (T0 . at time T0 . a bond may not only pay its principal back at maturity.5. Ti ). This income stream is equivalent to owning one Tn bond and kδ units of each Ti bond. Suppose a bond makes n regular payments at (uncompounded) rate k at times Ti = T0 + iδ (i = 1. but also make smaller regular coupon payments of a ﬁxed amount c up until then. . One interesting case is where the interest paid over an interval from time S to time T is the same as the yield of the T bond bought at time S . we use it to buy a T bond (or invest it in the bank account process Bt till time T ). as each payment is made. Instead. Especially at the long end. Ti ) −1 . Tn ) . We could instead roll up the payments into savings as we get them. but depended on current interest rates. . pure discount bonds with no coupon are not popular products. .
The payment made is
δL(Ti−1 ) = 1 P (Ti−1 − Ti ) − 1. Ti ) is known with respect to the FTi−1 information. then the ith payment will be determined by the δ period LIBOR rate set at time Ti−1 . Tn ) +
i=1
P (T0 . only the net difference is exchanged. we swap a ﬂoating interest rate for a ﬁxed one.
Using the tower law. Then the swap looks like a portfolio which is long a ﬁxed coupon bond and short a variable coupon bond. and buy some T2 bonds with
the dollar principal
• repeat until we are left with the dollar at time Tn . as shown in ﬁgure 5. n). is because the bond is equivalent to this simple sequence of trades:
• take a dollar and buy T1 bonds with it • take the interest from the bonds at T1 as a coupon. and we are left with exactly the right payoff at time Ti . we can divide it through both sides to get −1 −1 EQ BTi P −1 (Ti−1 . . That is. we might offer a contract where we receive a regular sequence of ﬁxed amounts and at each payment date we pay an amount depending on prevailing interest rates. INTEREST RATES
−1 −1 Because the conditional expectation EQ (BTi FTi−1 ) is BTi−1 P (Ti−1 . Ti ) and the bond price P (Ti−1 .
which is just P (T0 . . Ti ) = 1. In practice. Typically. Ti−1 ) − P (T0 .
Swaps
This very popular contract simply exchanges a stream of varying payments for a stream of ﬁxed amount payments (or vice versa).
n
V0 = P (T0 .
Suppose the swap pays at a ﬁxed rate k at each time period. Ti ) FTi−1 = BTi−1 . Ti−1 ) − P (T0 . If the payment dates are Ti = T0 + iδ (i = 1. Why this is so. . That is. Ti ). the bond has a ﬁxed price equal to the face value of its principal.130
CHAPTER 5. we can rewrite the value of the Ti payment as
−1 −1 BT0 EQ BTi−1 − BTi FT0 .9: A standard deﬁnition of the variable payment is that of the interest paid by a bond over the previous time period.
Surprisingly. so the initial prices must match. We
. Ti ) units of the Ti bond.
This has exactly the same cash ﬂows as the variable coupon bond. This price also suggests the hedge of selling a Ti bond and buying a Ti−1 bond. The total value of the variable coupon bond is the sum of its components. When the Ti−1 bond matures. . we buy P −1 (Ti−1 .
. T0 ). Ti ) is the forward price at time t for purchasing a Ti bond at time T0 That is Ft (T0 . Ti ) i=1
This rate k is the forward swap rate. . In this form the expression resembles the instantaneous swap rate. Tn ) . Tn ) + kδ
i=1
P (T0 . . we have chosen to receive ﬁxed payments at rate k . Ti ) = P (t. δ n P (T0 . Ti ).9:
know that the former is worth
n
P (T0 .
The ﬁxed rate needed to give the forward swap initial null value at time t is
k= P (t. Tn ) + kδ
i=1
P (T0 . The value of this swap at time T0 will be
n
X = P (T0 . The ﬁxed rate needed to give the swap initial null value is
k= 1 − P (T0 .6. Ti ) i=1
Forward swaps
In a forward swap agreement.
and the latter costs a dollar. Ti ) − P (t. δ n P (t. Ti ) i=1
where Ft (T0 . An alternative formulation of this expression is
k= 1 − Ft (T0 .
. n). . T0 ) − P (t. Tn ) . Tn ) + kδ i=1
P (t.
The present value of X at time t before T0 is given by the formula:
n −1 Vt = Bt EQ (BT0 XFt ) = P (t. Tn ) . T0 ). starting at time T0 with payments at times Ti = T0 + iδ (i = 1. δ n Ft (T0 . Ti ) − 1. Ti )/P (t.5. INTEREST RATE PRODUCTS
131
(a) Gross payments received and given
(b) Net receipts
Figure 5.
As we’ve seen before the value of this bond at time t is
n
Ct = P (t. . the formula is the same.
where r is the constant interest rate and F is the current forward price of the stock. t) F Φ log F + 1 σ 2 t k √ 2¯ σ t ¯ − kΦ
1 log F − 2 σ 2 t k √ ¯ σ t ¯
. struck at price k with exercise time t. has current worth
−1 EQ Bt P (t. T ). transferring the ownership of future (but not past) coupons along with it. as long as the other variables are expressed in terms of forward prices and term volatilities. T )/P (0.
where F is the current forward price for P (t. Under the Ho and Lee model. This makes the T bond and the discount bond lognormally distributed. An option on a T bond.
the forward rates and the instantaneous short rate are normally distributed. and under variable interest rates it is just the price of a tbond P (0. It is worth
V0 = e−rt FΦ log F + 1 σ 2 t k √ 2 σ t − kΦ log F − 1 σ 2 t k √ 2 σ t . with volatility σ . . T ) = σdWt + σ 2 (T − t)dt. so that we can price the option with the lognormal results of section 6. t). T )). which is the most general singlefactor model with lognormal bond prices. . n) before redeeming a dollar at time Tn . INTEREST RATES
Bond options
Like a stock option. struck at k with exercise time t.
where Q is the martingale measure. We can buy or sell the bond before time Tn . Ti ). The option price is
V0 = P (0. . T ) − k +
. and σ is the (term) volatility of St . Under constant interest rates this was ert . Otherwise.
Options on coupon bonds
Imagine a bond which pays coupons at rate k at the times Ti = T0 + iδ (i = 1. that is F = P (0. t).132
CHAPTER 5. a bond option gives the right to buy a bond at a future date for a given price. where the forward rates evolve as
dt f (t.
. but a different σ ¯ depending on the deterministic processes σt and φt in the model. this formula also holds with the same forward price.2. Compare this with the price of an option on a stock S . Tn ) + kδ
i=I(t)
P (t. that is F = ert S0 . Under ¯ ¯ the Vasicek model. We see that the bond option price formula merely changes the discount factor representing the value now of a dollar at time t. σ 2 t is the logvariance of P (t. and the term volatility σ is σ(T − t) (that is.
T ) can be seen as a deterministic function P (t. Ti ) − Ki
+
. rt ) which is decreasing in rt . Ti ) is larger than Ki . Setting Ki to be P (t. Ti . In general it is not easy to value this option analytically.6. which equals
−1 (1 + kδ)Bt EQ BTi−1 (K − Pi )+ Ft .
. we can price the cap. . in the special case where we have a singlefactor model with a Markovian short rate. Each bond price P (t.
Caps and ﬂoors
Suppose we are borrowing at a ﬂoating rate and want to insure against interest payments going too high. and we can price each one using the zerocoupon bond option formula. However. then we pay at time Ti the δ period LIBOR rate set at time Ti−1
L(Ti−1 ) = 1 δ 1 P (Ti−1 − Ti ) −1 .
In other words. INTEREST RATE PRODUCTS
133
where I(t) = min{i : t < Ti } is the sequence number of the next coupon payment after time t. An individual payment at a particular time Ti is called a caplet. exercised at Ti−1 . maturity and the instantaneous rate.
where Pi is P (Ti−1 . Ti ) and K is (1 + kδ)−1 . Suppose we have an option to buy the bond at time t for price K . This means that Ct is larger than K if and only if any (and every) P (t. .5. then r∗ is also critical for an option on the Ti bond struck at Ki .
This is just equal to the value of (1 + kδ) put options on the Ti bond.
How much would it cost to ensure that this rate is never greater than some ﬁxed rate k ? The cap contract pays us the difference between the LIBOR and the cap rate
δ L(Ti−1 ) − k
+
at each time Ti . A portfolio which is long a number of bonds will have the same behavior. . So Ct itself will be a function C(t. Tn ) − Kn
+ + n
+ kδ
i=I(t)
P (t. rt ) of time. Now we can rewrite the caplet claim as
X = (1 + kδ)Pi−1 (K − Pi )+ . The value of the caplet at time t is −1 Bt EQ (BTi X mcFt ). and if we can price caplets. we can price the option more easily using a trick of Jamshidian. Additionally. Thus there is some critical value r∗ of r such that C(t. prices fall. r∗ ) is exactly K . The option price formula (and putcall parity) will then price the caplet. n). this function will be decreasing in rt — as rates rise. If we make payments at times Ti = T0 + iδ (i = 1. an option on this portfolio is a portfolio of options. r∗ ). And so
(Ct − K) = P (t. . T . struck at K .
λ
. Tn ) + kδ
i=1
P (T0 . INTEREST RATES
A ﬂoor works similarly.
where σ1 . . The HJM completeness conditions reduce. but inversely.
Swaptions
A swaption is an option to enter into a swap on a future date at a given rate. a swap option looks like a bond option. σ2 and λ are constants. There is a ﬂoorcap parity which says that the worth of a ‘ﬂoorlet’ less the cost of a caplet equals (1 + kδ)P (t. . so we can read off information about it from that structure. to there being two F previsible processes γ1 (t) and γ2 (t) such that the drift α is
2 α (t. The swap payment dates are Ti = T0 + iδ (i = 1. That is. whereas W2 gives shortterm shocks which have little effect on the longterm end of the curve. n). and the ﬁxed swap rate is k . on a Tn bond which pays a coupon at rate k at each time Ti . This model is HJM consistent. T ) = σ1 dW1 (t) + σ2 e−λ(T −t) dW2 (t) + α(t. Ti ) − P (t. and α is a deterministic function of t and T . T ) = σ1 γ1 (t) + σ2 e−λ(T −t) γ2 (t) + σ1 (T − t) + 2 σ2 1 − e−λ(T −t) e−λ(T −t) . you have a long position in the bond market.
This is exactly the same as a call option. If you receive ﬁxed on a swap. in that we receive a premium for agreeing to never pay less than rate k at each time Ti .134
CHAPTER 5. Suppose we have an option to receive ﬁxed on a swap starting at date T0 . Then the worth of the option at time T0 is
n +
P (T0 .
5. A simple case is given in HeathJarrowMorton’s original paper. struck at 1. T )dt.7
Multifactor models
If we want to price a product depending on a range of bonds. Buying a ﬂoor and selling a cap at the same strike k is exactly equivalent to receiving ﬁxed at rate k on a swap. That is not entirely a coincidence as a swap is just a coupon bond less a ﬂoating bond (which always has par value). It is an extension of Ho and Lee’s model to two factors. Ti−1 ).
A twofactor model
Suppose the forward rates evolve as
dt f (t. we pay an extra amount
δ k − L(Ti−1 )
+
at time Ti . in this case. it makes more sense to use a multifactor model. . . Ti ) − 1
. Here the W1 Brownian motion provides ‘shocks’ which are felt equally by points of all maturities on the yield curve.
the short rate loses its dominant role as the carrier of all information about the bond prices.
0
Like Ho and Lee. However. T ) k 2¯ σ (t. MULTIFACTOR MODELS
135
So the range of available drifts has two degrees of functional freedom away from the martingale measure drift.
The general multifactor normal model
We can actually generalize the twofactor model above to a general multifactor one which also has normal forward rates and an explicit BlackScholes type option pricing formula. Bt = exp
t 0 rs ds
. T ) = σ1 (T − t)2 t + ¯
σ2 1 − e−λ(T −t) λ
2
1 1 − e−2λt . T ) k 2¯ σ (t. T ). exercised at time t is
V0 = P (0.
. This BlackScholes type of formula allows us to price caps and ﬂoors as well as options on the discount T bonds. We can deduce from the forward T rate formula that − log P (t. T ) = xi (t)yi (T ). the trick we used before to price options on couponbearing bonds does not work. Nevertheless the model does have the advantages of technical tractability and an explicit option formula. T ) can be written as a product
σi (t. T ) = t f (t. where each volatility surface σi (t. u) du +
t 0 t
α (s. T ) ds. struck at k .
This means that the instantaneous rate is made up of a Brownian motion and an independent meanreverting (OrnsteinUhlenbeck) process plus drift. Setting σ 2 (t. t). t) F Φ log F + 1 σ 2 (t. making it more involved to price them and the associated swaptions. T ) ¯ . given the joint lognormality of the asset and discount bond prices. T )/P (0. u) duds. is also lognormally distributed.2. T ) to be the variance (term variance) of log P (t. the forward rate is
f (t.7. T ) ¯ − kΦ log F − 1 σ 2 (t. u) du is
σ2 −λt σ1 (T − t) W1 (t)+ e − e−λT λ
t 0
e dW2 (s) +
λs
T
t
T
f (0. We take the instance of the completely general nfactor model. We can use the results of section 6. T ) +
t
α (s. The value of an option on the T bond. However in a multifactor setting. the forward price of the T bond.5. Under the martingale measure (that is γ1 = γ2 = 0). this model has normally distributed forward rates — which does allow them to go negative. 2λ
The discounted bond. we have ¯
2 σ 2 (t. T ) = σ1 W1 (t) + σ2 e−λT
t 0
eλs dW2 (s) + f (0. because
we can deduce that the integral 0t rs ds is normal from the expression for rt above.
where F is P (0. in the multifactor setting.
γn . For the market to be complete. and let σ be the term volatility of the T bond up to time t. there should be n F previsible processes γ1 . T ) =
i=1 T
xi (t) yi (t) γi (t) + xi (t) Yi (t. P (t. for instance. t) F Φ log F + 1 σ 2 t k √ 2 σ t − kΦ
1 log F − 2 σ 2 t k √ σ t
. INTEREST RATES
where xi and yi are deterministic functions. The forward rates are then driven by
n
dt f (t.136
CHAPTER 5. T ) = σi (t) exp − λi (T − t) .
Here the function xi determines the size at time t of ‘type i shocks’. Consequently the bond prices are lognormally distributed and a BlackScholes type formula holds (see section 6. Firstly. exercisable at time t is
V0 = P (0. this framework incorporates both the Ho and Lee model (x(t) = σ . T ) is the δ period (forward) LIBOR rate for borrowing at a time T . . where aij (t) = Yj (t. T ) −1 . T )dt. the short rate and the forward rates are normally distributed. struck at k . T ). we need two conditions on the functions α and yi to hold. T ) .
. For the general volatility surface σi (t. F = P (0. T )/P (0. We shall simplify their notation slightly and write
L (t. T ) = xi (t)yi (T ). . the drifts consistent with hedging span an ndimensional function space around the martingale drift. T ) =
i=1
yi (T )xi (t)dWi (t) + α(t. Ti ) should be nonsingular for all t < T1 .2).
where the σi (t) are deterministic functions of time and the λi are distinct constants. for every set of n maturities T1 < · · · < Tn . that is σ 2 t is the variance of log P (t. T + δ)
So L(t. T )
i=1
t 0
x2 (s)ds. y (T ) = exp − 0 αs ds ). i
Then the value at time zero of a call on the T bond.
BraceGatarekMusiela
The BraceGatarekMusiela (BGM) model is a particular case of HJM which focuses on the δ period LIBOR rates. Let F be the forward price of the T bond at time t. Secondly the matrix At (aij (t)). and the function yi controls how the shock is felt at different maturities. y(T ) = 1) t T and the Vasicek model (x (t) = σt exp 0 αs ds . or
1 σ = t
2 n
Yi2 (t. T ) = 1 δ P (t. It is satisﬁed. if each volatility σi has the form
σi (t.
where Yi (t. This condition is really just asserting that all the functions yi are different. In other words. In the singlefactor case when n = 1. t). . T ) = t yi (u)du. such that
n
α (t. .
Ti ) F Φ log F + 1 ζ 2 (t. T ) = L (t. . Ti−1 ) and ζ 2 (t. we only need to know the function γ . T )
holds for all t less than T . T )2 ds. Writing L(T ) for L(T. T )
i=1
γi (t. equal to
δP (t. is not only a PT +δ martingale. We shall see later that this enables us to price caps and swaptions easily. Brace. For instance. T ) 1 + δL (t. suppose we have a contract which pays off at a sequence of times Ti = T0 + iδ (i = 1. for example if X = f (L(Ti )). Ti−1 )
T
. in their paper. This valuation has the familiar BlackScholes form because under the forward measure PTi . T ) is restricted in the BGM setup to those σ such that
T +δ T
σi (t. In this case. One such simple f is the caplet payoff δ(L(Ti−1 ) − k)+ at time Ti . MULTIFACTOR MODELS
137
The general HJM model (of n factors) deﬁned by the forward volatilities σi (t.7. the worth of the caplet at time t is Vt . γ is some deterministic Rn valued function which is absolutely continuous with respect to T . T )
i=1
˜ γi (t. Then it follows that. T ) = f (t)γi (T − t)
by calibrating against known prices of caps and swaptions. the instantaneous LIBOR rate. under the forward measure PT +δ (see section 6.
˜ where Wi are PT +δ Brownian motions. Ti−1 ) k 2 ζ (t. u) du dt . in practice γ is calibrated by comparing the model’s prices with the market. then the value of that payment at time t is
Vt = P (t. but is also lognormal. T ) is t γ(s.
More interestingly. as a tprocess. T ) dWi (t).
where F is the forward LIBOR rate L(t.
. . L(Ti−1 ) is lognormal and the calculation proceeds as usual. T ) = L (t. Ti−1 ) − kΦ log F − 1 ζ 2 (t. under the martingale measure Q. If the payment at time Ti+1 depends on the LIBOR rate set at time Ti . While the γ function represents the correlation at time t between changes in the LIBOR rates at different forward dates T . n). L obeys
n
dt L (t. u) du =
δL (t. the variance of log L(T ) given Ft . Gatarek and Musiela ﬁt a γ function of the form γi (t. . T ) dWi (t) +
t
σi (t. T ).4). Here. Ti−1 ) k 2 ζ (t. T ) γi (t. rather than the whole volatility structure. L obeys the SDE
n T +δ
dt L (t.5. Ti+1 )EPTi+1 f (L(Ti ))Ft . T ). allows us to evaluate this expression for simple f .
The fact that L(Ti ) is lognormally distributed under PTi+1 . To price. . Thus L(t.
. We also deﬁne
i
di =
j=1
δL (t. Tj−1 ) −1 (1 + δL (t. . .138
CHAPTER 5. Ti−1 ) given Ft under the forward measure PTi .
which is the variance of log L(T0 . Tj−1 ) exp (Γj (s + dj ))) = 1.
and s0 to be the unique root of the equation
n
(kδ + I (i = n))
i
s:
i=1
j=1
Then an approximation to the value at time t of the above swaption is
n
Vt = δ
i=1
P (t. Ti ) L (t. Consider the option to pay ﬁxed at rate k and receive ﬂoating and at times Ti = T0 + iδ (i = 1. 2 1 + δL (t.
where Fi = −Γi (s0 + di ).
. Let us set
Γ2 = i
T0 t
γ (s. n). . Tj−1 ) Γj − 1 Γi . Ti−1 )2 ds. Ti−1 ) Φ
Fi + 1 Γ2 2 i Γi
− kΦ
Fi − 1 Γ2 2 i Γi
. INTEREST RATES
We can even (approximately) price swaptions.
Also we will reveal the underlying framework which governs all these models from behind the scenes. t).
Here r is the constant interest rate. σ is the constant stock volatility and µ is the constant stock drift. It assumes that there is only a single stock in the market.4. or even a function of both the stock price and time σ(St .1 General stock model
We recall that the BlackScholes model contained a bond and a stock Bt and St with SDE s
dBt = rBt dt. and we are using the SDE formulation discussed in section 4. Not only can they vary with time.) We will replace a by a general F previsible process σt . on movements of the Brownian motion W ). None of these assumptions is necessary. (For instance the volatility at time t might depend on the maximum value achieved by the stock price up to time t. The process W is PBrownian motion.Chapter 6
Bigger models
T
he BlackScholes stock model assumes that the stock drift and stock volatility are constant. it is analyzable in this framework. And it assumes that the cash bond is deterministic with zero volatility. But if a model is driven by Brownian motions. This is not to say that all models. The subsequent sections tackle these restrictions one by one and show how a more general model can still price and hedge derivatives. no matter how complex or bizarre. Our most general stochastic process can have variable drift and volatility. and the constants r and µ by F previsible processes rt and µt respectively.
6.
139
.
and
dSt = St (σdWt + µdt). will always give good prices. Even this is not fully general.
and
dSt = St (σt dWt + µt dt). We could replace the constant σ by a function of the stock price σ(St ). and has no transaction costs. but they can depend on movements of the stock itself (or equivalently. The new SDEs are now
dBt = rt Bt dt.
Et = EQ BT XFt Then the martingale representation theorem (section 3. σt
as was adumbrated in the market price of risk section (4. Under Q. 0T rt dt. Z has the SDE
˜ dZt = σt Zt dWt .]
Change of measure
−1 As before. rt and µt cannot be ﬁlly general.
. continue to be independent of the instrument considered.) Let us take φt to be our stock portfolio holding at time t. (Note that we need σt never to be zero. t ˜ This is achieved by adding a drift γt to W . with maturity at time T . that γt satisﬁes the CMG growth condition EP exp 2 0T γt dt < ∞. We can form a Qmartingale Et through the conditional expectation process of the −1 discounted claim. we aim to make the discounted stock price Zt = Bt St into a martingale.
[Technical note: the processes σt . It will. For instance.
t 0 t
σs dWs +
0
1 2 (µs − 2 σs )ds . It should also be checked. then the procedure is not much different from the basic BlackScholes technique.5) says that the martingale Et is the integral
t
Et = E0 +
0
φs dZs . That is. the integrals 0T σt dt. and 0T µt dt are ﬁnite.140
CHAPTER 6.
And Z is a Qmartingale if
γt =
µt − r t . Then
dEt = φt dZt .
so it is at least a local martingale because it is driftless. then Zt has SDE
˜ dZt = Zt σt dWt + (µt − rt − σt γt ) dt . Explicitly. It should also be checked that 1 T 2 Z is a proper martingale. however. 1 2 in any actual case. we need that (with P2 probability one). Now the market price of risk depends on the time t and the sample path up to that time.
Replicating strategies
If X is the derivative to be priced. if Wt = Wt + 0 γs ds is QBrownian motion.
for some F previsible process φt .4). it is enough that EQ exp 2 0 σt dt is ﬁnite. as they must be integrable enough for these integrals to exist. BIGGER MODELS
These have solutions
t
Bt = exp St = S0 exp
0
rs ds .
2. To make speciﬁc calculations. then an approximation for φt or “dVt /dSt ” is the delta hedge
φt ≈ ∆Vt . even if we are not working with the BlackScholes model (as in section 4. If we can approximate the price Vt at time t.
In other words. then the value of the portfolio at time t is
Vt = φt St + ψt Bt = Bt Et .
Derivative pricing
Arbitrage arguments convince us that the only value for the derivative at time t is Derivative price
−1 Vt = Bt EQ BT XFt = EQ T
exp −
t
rs ds X Ft
. (The lognormal cases of section 6.2 Lognormal models
We have already seen that the BlackScholes formula can be true.
−1 So (φ. t + ∆t).2 will be notable exceptions. the value at time t is the suitably discounted expectation of the derivative conditional on the history up to time t. the volatility of the stock σt — though not its drift — and the derivative itself. these expectations cannot be performed analytically. if the model is much more complex than BlackScholes. There is no general expression which will provide a more explicit answer for the option value Vt . under the measure which makes the discounted stock process a martingale — the riskneutral measure.
6. ψ) is a selfﬁnancing strategy with initial value V0 = EQ BT X and terminal value VT = X .
. ψ) is selfﬁnancing in that the changes in the value Vt are due only to changes in the assets’ prices. one needs to know the discount rate rt . That is
dVt = φt dSt + ψt dBt . LOGNORMAL MODELS
141
Setting the bond portfolio holding ψt to be ψt = Et − φt Zt .
It also follows (as in chapter three) that (φ. The common feature of models where this happens is that the asset prices are lognormally distributed under the martingale measure Q.1).6.) Instead numerical methods must be used.
Implementation
In practice. ∆St
where ∆ represents the change over a small time interval (t.
Reexpressing ST :
1 2 ST = F exp α1 Z − 2 α1 − ρα1 α2 .
We can see why these formulae are true. under the martingale measure.142
CHAPTER 6. Write ST as
2 ST = A exp α1 Z − 1 α1 . for some currency and equity models. Explicitly. 1) independent of Z . The forward price to purchase F at time T is
F = S0 erT .
and the price of a call on ST struck at k is the generalized BlackScholes formula
−1 V0 = EQ BT
FΦ
2 log F + 1 σ1 T k √2 σ1 T
− kΦ
2 log F − 1 σ1 T k √2 σ1 T
. we get
−1 1 2 BT = B exp α2 (ρZ + ρW ) − 2 α2 .
. Then the forward price for purchasing S at time T is
F =
−1 EQ BT ST −1 EQ BT
. suppose the stock ST and the cash bond BT are known to be jointly 2 lognormally distributed under the martingale measure Q.
And the value at time zero of an option to buy ST for a strike price of k is
V0 = e−rT FΦ log F + 1 σ 2 T k √ 2 σ T − kΦ log F − 1 σ 2 T k √ 2 σ T .
where r. ¯
with
2 2 α2 = σ2 T. (σ1 and σ2 are term volatilities). σ2 T be the variance of log BT .
where A is the constant EQ (ST ) and Z is a normal N (0. ¯ The expected discounted stock price is then
−1 EQ BT ST = AB exp 1 2 2 2 2 (α1 + ρα2 )2 + 1 ρ2 α2 − 1 α1 − 1 α2 = AB exp (ρα1 α2 ) . 1) random variable under Q.
or equivalently
F = exp (ρσ1 σ2 T ) EQ (ST ) . 2¯ 2 2
So the forward price for ST is thus F = A exp(ρσ1 σ2 T ). and let ρ be their correlation. and also for simple interest rate models. are lognormal. there are great advantages.
where ρ = 1 − ρ2 and W is a normal N (0. Let σ1 T be the variance of −1 2 log ST . BIGGER MODELS
In the simple BlackScholes model. the cash bond and the stock are modeled as
Bt = ert
and
St = S0 exp (σWt + µt) . σ and µ are constants and W is PBrownian motion. 2 The discount factor is lognormal with logvariance σ2 T and its correlation with the −1 stock logprice is ρ. This holds for the BlackScholes model itself. Setting B to be its expectation B = EQ (BT ).
Lognormal asset prices
When prices. 2
with
2 2 α1 = σ1 T.
Stochastic processes adapted to ndimensional Brownian motion A stochastic process X is a continuous process (Xt : t ≥ 0) such that Xt can be written as n
t
Xt = X0 +
i=1
0
i σi (s)dWs +
t
0
µs ds. A good model of several securities must not only describe each one individually. more complex equity products. Z −z = Φ (y + z). having modeled its behavior adequately. where IA is the indicator function of the event A. General Motors stock. where a swap’s current value is affected by the movements of a large number of bonds of varying maturities. but also represent the interaction and dependency between them. . n independent Brownian
. for any constants y and z . The differential 0 form of this equation can be written
n
dXt =
i=1
σi (t) dWti + µt dt. .
Multiple stocks can be driven by multiple Brownian motions. If we write an option on.3 Multiple stock models
BlackScholes assumes a single stock in the market. in the nfactor case. Instead of just one PBrownian motion. [The notation E(X.Z
−z .6. MULTIPLE STOCK MODELS
143
gives us the call value
−1 V0 = EQ BT (ST − k)+ = BEQ eρα2 Z− 2 ρ
1 2 2 α2
(ST − k)+ . we will have. In many cases.
which is also equal to
BEQ F e(α1 +ρα2 )Z− 2 (α1 +ρα2 ) − keρα2 Z− 2 ρ
1 2 1 2 2 α2
. this assumption does little harm.]
6. we are unaffected by the movements of other securities. σn and µ are random F previsible processes such that the integral t 2 i σi (s) + µs  ds is ﬁnite for all times t (with probability 1). say. or equivalently is E(XIA ). These two processes have some degree of codependence. our quanto contract of section 4. Using the probabilistic k 1 2 result that E eyZ− 2 y . . In particular. large movements in one may be linked with corresponding movements in the other. Even more so in the bond market.5 was related to both the sterling/dollar exchange rate and an individual UK stock. For instance. depend on the behavior of at least two separate securities.
1 2 where z is the critical value z = log F − 2 α1 − ρα1 α2 /α1 . Such changes would suggest that the two securities are correlated. .
where σ1 . the result follows. A) denotes the expectation of the random variable X over the event A.3. such as quantos. However.
o Itˆ ’s formula (nfactor) o If X is a stochastic process. There is also an nfactor version of Itˆ ’s formula and the product rule. . and Y is a stochastic process satisfying dYt = i ρi (t) dWti + νt dt. . . If however Xt and Yt are adapted to two independent Brownian motions. The total 2 2 volatility of the process X is σ1 (t) + · · · + σn (t). and f is a deterministic twice continuously differentiable function. . If Xt and Yt are both adapted to the same Brownian motion Wt .144
CHAPTER 6. then Yt := f (Xt ) is also a stochastic process with stochastic increment
n n
dYt =
i=1
σi (t) f (Xt ) dWti +
µt f (Xt ) +
1 2 i=1
2 σi (t) f (Xt ) dt. . In other words. satisfying dXt = i σi (t) dWti + µt dt. then Xt will have zero volatility with respect to W 2 . the variance of 2 2 dXt is i σi (t)dt. That means that each Wti behaves as a Brownian motion. then this rule agrees with the ﬁrst case.
This new version uniﬁes the two apparently different cases of the product rule we encountered in section 3. ρ1 (t) = 0. Their ﬁltration Ft is now the total of all the histories of the n Brownian motions. . but there is now a volatility process σi (t) for each factor. Wtn .
Again this is an analogue of the onefactor Itˆ formula.3. but strictly is now a vector. This leads to an enhanced deﬁnition of a stochastic process (see box). Thus the term σi (t)ρi (t) in the nfactor product rule will be identically zero. say Wt1 and Wt2 . .3. that is σ2 (t) = 0. and similarly Yt will have zero volatility with respect to W 1 . FT is the history of the ndimensional vector (Wt1 . made up of the contribution σi (t)dt from each Brownian motion component W i . The CameronMartinGirsanov theorem continues to hold where W is ndimensional 1 T Brownian motion and the drift y is an nvector process for which EP exp 2 0 γt 2 dt
. The drift term is unchanged from the original (onefactor) deﬁnition. Wtn ) up to time T . BIGGER MODELS
motions Wt1 . agreeing with the second case in section 3. In other words. . with the replication of the o volatility terms for each additional Brownian factor. then Xt Yt is a stochastic process satisfying
n
d (Xt Yt ) = Xt dYt + Yt dXt +
i=1
σi (t) ρi (t) dt. Product rule (nfactor) If X is a stochastic process satisfying dXt = i σi (t) dWti + µt dt. We must remember that in a multifactor setting volatility is no longer a scalar. the variances adding because the Brownian motion components are independent. and the behavior of any one of them is completely uninﬂuenced by the movements of the others.
The situation is not quite as simple as that (the bond market. . n. With W as ndimensional QBrownian motion. γt ) is an F previsible nvector process which satisﬁes the growth t i 1 T ˜ condition EP exp 2 0 γt 2 dt < ∞. and n different market n 1 securities St .
There is also a converse to this theorem. φn ) such t t that
n t 0
Nt = N0 +
j=1
j φj dMs . . St . . In integral
. Their SDEs are
dBt = rt Bt dt. .j=1 of processes. exactly analogous to the one factor converse. has an unlimited number of different maturity bonds). . and N any other onedimensional Qmartingale.3. . . and t (σij )n is its volatility vector. CameronMartinGirsanov theorem (nfactor) Let W = (W 1 . Our model then. the collection j=1 of n such vectors forms a volatility matrix Σt = (σij (t))n i. . will contain a cash bond Bt as usual. . . W n is ndimensional QBrownian motion up to time T . .6.5 that there is an nfactor martingale representation theorem. . .
i i dSt = St j=1 n
σij (t) dWti + µi dt . for instance. . µi is the drift of the ith security. equivalent to P up to time T . W n ) be ndimensional PBrownian motion. then there is an F previsible nvector process φt = (φ1 . . we recall from section 5. and we set Wti = Wti + 0 γs ds. .
Here rt is the instantaneous shortrate process. t i = 1. s
The general nfactor model
We will see later that it is important that we have essentially as many basic securities (excluding the cash bond) as there are Brownian factors. . . Then there
˜ ˜ ˜ is a new measure Q. Suppose that 1 n γt = (γt . but we shall start with the canonical case. . Generally speaking. Finally. MULTIPLE STOCK MODELS
145
is ﬁnite. M as an ndimensional Qmartingale with nonsingular volatility matrix. and if there are fewer we will not be able to hedge. The RadonNikodym derivative of Q by P is dQ = exp − dP
n i=1 0 T i γt dWti − 1 2 T 0
γt 2 dt . . . if there are more securities than factors there might be arbitrage. . . As each security has a volatility vector. . such that W := W 1 .
. EQ exp 1 n 0 σij (t) dt < ∞. n. . though. 1. µt and rt .146
CHAPTER 6. Then the discounted stock −1 i price Zti = Bt St has SDE
i i dZt = Zt j=1 n
˜ σij (t) dWtj + µi − rt − t
n j=1
j σij (t) γt dt . then there is a measure Q which makes the discounted stock prices into Qmartingales. We T 2 also need the integral condition that for each i. under which all the discounted stock prices are Qmartingales simultaneously. . t
−1 The onefactor market price of risk formula γt = σt (µt − rt ) is now just a special case. . . by the nfactor CMG theorem. we must have that
n i σij (t) γt = µi − rt . t j=1
for all t. BIGGER MODELS
form. . this can be reexpressed as
Σt γt = µt − rt 1. . then a unique such γt must exist and be equal to
γt = Σ−1 (µt − rt 1) . If.
In terms of vectors and matrices. . This vector equation may or may not have a solution γt for any particular t.)
. the matrix Σt is invertible. 1).
where Σt is the matrix σij (t) and 1 is the constant vector (1. This means that if Σt is invertible for every t and γt satisﬁes the CMG 1 T condition EP exp 2 0 γt 2 dt < ∞. .
To make the drift term vanish for each i. these securities are
t
Bt = exp
i i St = S0 exp
0
rs ds . . 1 n Suppose we add a drift γt = (γt .
i = 1. (Or at least into Qlocal martingales. Whether it does or not depends on the actual values of Σt . . . γt ) to Wt . so that
˜ Wti = Wti +
t 0 i γs ds
is QBrownian motion.
n j=1 0
t j σij (s)dWs + t 0
µ i − 1 s 2
n j=1
2 σij (s) ds
Change of measure
We now want to ﬁnd a new measure Q. for j 2 Z i to be a proper Qmartingale.
But we do have to check that the selfﬁnancing equations will still work. φn .6. We have seen in the foreign exchange context that there can be a choice of which currency’s cash bond to use. or equivalently dEt = φt dZt . We want to show that Selfﬁnancing strategies A portfolio strategy (φt . . But no matter which numeraire is chosen. . t
so that the value of the portfolio is Vt = Bt Et . . we assumed that the numeraire had no volatility. t the bond holding ψ is
n
ψt = Et −
j=1
j φj Zt . . ψt ) of holdings in a stock St and a possibly volatile cash bond Bt has value Vt = φt St + ψt Bt and discounted value Et = φt Zt + ψt . It is because the choice of numeraire doesn’t matter. s
The invertibility of Σt is essential at this stage. . t
'
$
Derivative pricing The value of the derivative at time t is
−1 Vt = Bt EQ BT XFt = EQ T
exp −
t
rs ds X Ft
.4 Numeraires
Although the numeraire is usually chosen to be a cash bond. . ψt ) t t where φi is the holding of security i at time t and ψt is the bond holding. and let Et be the Qmartingale Et = −1 EQ BT XFt If the matrix Σt is always invertible.4. In fact. As usual. The portfolio is selfﬁnancing in that
n
dVt =
j=1
j φj dSt + ψt dBt . then the nfactor martingale representation theorem gives us a volatility vector process φt = (φ1 . When we proved the selfﬁnancing condition in chapter three. Then the strategy is selfﬁnancing if either dVt = φt dSt + ψt dBt . . that we usually pick the stolid cash bond. . φn ) such that t t
n t 0
Et = E0 +
j=1
j φj dZs . the price of the derivative will always be the same. it needn’t be.
%
&
6. it can be any of the tradable instruments available. not only can the numeraire have volatility. Our hedging strategy will be (φ1 .
. NUMERAIRES
147
Replicating strategies
Let X be a derivative maturing at time T . where Z −1 is the discounted stock process Zt = Bt St . This is not actually necessary.
. n)
and
−1 Ct Bt
are QC martingales. . St and two others Bt and Ct either of which might be a numeraire. with discounted value Et satisfying dEt = φt dZt . .
where X and Y are stochastic processes with stochastic differentials
dXt = σt dWt + µt dt. ψ) is selfﬁnancing. . We can use the substitutions dEt = φt dZt and Et = φt Zt + ψt to rearrange the above expression into
dVt = φt (Bt dZt + Zt dBt + σt ρt dt) + ψt dBt . ψ).
Changing numeraires
1 n Suppose we have a number of securities including some stocks St .
ζs EQC (Xt Fs ) = EQ (ζt Xt Fs ) . . . n) and Bt Ct
are Qmartingales.
The second use of the product rule says that the term in brackets above is equal to d(BZ)t = dSt . or at least what its RadonNikodym derivative with respect to Q is.
Suppose we have a strategy (φ. . We want to show that (φ. . we need to ﬁnd a measure Q (equivalent to the original measure) under which −1 i −1 Bt St (i = 1. . . Then we would have a different measure QC under which
−1 i Ct St
(i = 1. We can actually ﬁnd out what QC is. Then the value at time t of a derivative payoff X at time T is
−1 Vt = Bt EQ BT XFt . It follows from this
.
dQC dQ
Ft .148
CHAPTER 6. Firstly
dVt = d (Bt Et ) = Bt dEt + Et dBt + σt (φt ρt ) dt.
Suppose however that we choose Ct to be our numeraire instead. dYt = ρt dWt + νt dt. The resulting equation is the selfﬁnancing equation. .
where σt is the volatility of Bt and ρt is the volatility of Zt (and hence φt ρt is the volatility of Et ). This also holds for nfactor models with multiple stocks. If we choose Bt to be our numeraire.4. We recall RadonNikodym fact (ii) from section 3. BIGGER MODELS
Recall the onefactor product rule
d (XY )t = Xt dYt + Yt dXt + σt ρt dt.
where ζt is the change of measure process ζt = EQ that if Xt happens to be a QC martingale. then
ζs Xs = EQ (ζt Xt Fs ) . We do this with two applications of the product rule. that for any process Xt . .
Calculating the forward price for X is now only a matter of taking its expectation under the forward measure. Ct St . T ). Bt P (0. as well as the δ period LIBOR rate for borrowing up till time T . . If we call this numeraire Ct . Once more. T ) a PT martingale.
dQC CT = . it is often popular to use a bond maturing at date T (the T bond with price P (t. T ). The forward measure PT thus has RadonNikodym derivative with respect to Q of
dPT CT 1 − − . Bt St . The canonical QC martingales (including the constant martingale with value 1) are −1 −1 1 −1 n −1 −1 1 −1 n 1. just as in the foreign exchange section (4. then
−1 −1 −1 VtC = ζt Ct EQ ζT CT XFt = Bt EQ BT XFt .1). Ct Bt . T )/P (0. 1. then Ct = P (t. T ) BT
The associated Qmartingale is
ζt = EQ dPT Ft dQ = Ct P (t. . T ) dWi (t). namely Ft = P −1 (t. . where the dollar and sterling investors agreed on all derivative prices. Bt St . dQ BT
The price of a payoff X maturing at T under the QC measure is
−1 VtC = Ct EQC CT XFt . . T ) = .4. T )) as the numeraire.
. so the two agree. . NUMERAIRES
149
and so ζt Xt is a Qmartingale. −1 Each corresponding pair has a common ratio of ζt = Bt Ct .
Using again the RadonNikodym result that EQC (XFt ) = ζt−1 EQ (ζT XFt ).
This is exactly the same as the price Vt under Q.6. . by property (ii) of the RadonNikodym derivative. T ) Bt
Now the forward price set at time t for purchasing X at date T is its current value Vt −1 scaled up by the return on a T bond.
so is itself a PT martingale. Ft equals
Ft = EPT (XFt ) . The new numeraire is the T bond normalized to have unit value at time zero. Thus the RadonNikodym derivative of QC with respect to Q is the ratio of the numeraire C to the numeraire B . T ) Bt EQ BT XFt . . we ﬁnd that ζt satisﬁes
n
dζt = ζt
i=1
Σi (t. .
Example — forward measures in the interestrate market
In interestrate models. The martingale measure for this numeraire is called the T forward measure PT and makes the forward rate f (t. dQ BT P (0. Ct St and similarly the Qmartingales are Bt Ct . From the SDE for P (t.
. so that f (t. the volatilities σ .
n
and
dt f (t. plus a currency market linking them. Now we will. T ) is a PT martingale with
f (t. . Wtn . T ) are the forward rates for rT .
. τ and ρ are nvectors σi (t.
Another forward measure martingale is the δ period LIBOR rate
Lt = 1 δ P (t. What we have here are two separate interestrate markets (the dollar denominated and the sterling denominated). . T ) and ρi (t) (i = 1. Of course n might be one. T )
See chapter ﬁve (section 5. but it needn’t be. . Also the forward rates f (t. we will imagine ourselves to be a dollar investor operating in both the dollar and sterling interestrate markets. For deﬁniteness. By the converse of the CMG theorem. T ) = EPT (rT Ft ) .5 Foreign currency interestrate models
We have looked at foreign exchange (section 4. . τi (t. then its value at time t is
−1 Vt = Bt EQ BT XFt = P (t. T ). T − δ) −1 . The multifactor model approach is needed to reﬂect varying degrees of correlation between various securities in the three markets. we will work in an nfactor model driven by the independent Brownian motions Wt1 . T ) dWi (t) . If X is a payoff at date T . P (t. T ) with respect to Wi (t). . Our variables will be: As in the HJM model.150
CHAPTER 6. n). We have looked at the interest rate market (chapter ﬁve).
So the value of X at time t is just the PT expectation of X up to time t (the forward price of X ) discounted by the (T bond) time value of money up to date T . But we have not yet studied an interest rate market of another currency. T ) ds
is PT Brownian motion. .
6. and Σi (t. T ) EPT (XFt ) .7) for more details. in which case. T ) =
i=1
˜ σi (t. BIGGER MODELS
where W is ndimensional QBrownian motion. we see that
˜ Wi (t) = Wi (t) −
t 0
Σi (s. T ) is the component of the volatility of P (t. . This gives an alternative expression for pricing interestrate derivatives.1).
T ). and let the dollar discounted value of these three securities be X . T ) is the W i volatility term of P (t. T ) the drift of f (t. T ))
the volatility of g(t. u) du. T ) . T ) . Ti and Ti . τi (t. T ) dt. Ti (t.
Then Σi (t. t)) the sterling cash bond (equal to exp
t 0
us ds)
the exchange rate value in dollars of one pound the logvolatility of the exchange rate the drift coefﬁcient of the exchange rate (the drift of dCt /Ct ). T ) the sterling short rate (equal to g(t. the dollar tradable securities in this market consist of the dollarbonds P (t.
˜ It will simplify later expressions to introduce the notation Σi . T ) : σ(t. and the dollar worth of the sterling cash bond Ct Dt . Let us ﬁx T . T ) = Ti (t. T ) : β(t. T ) : α(t. T ) the drift of g(t. T ) = −
t T t
σi (t.
i=1
Apart from the dollar cash bond Bt .1: Notation P (t. the dollar worth of the sterling bonds Ct Q(t. FOREIGN CURRENCY INTERESTRATE MODELS
151
Table 6. T ) : ut : Dt : Ct : ρt : λt : the dollar zerocoupon bond market prices
∂ the forward rate of dollar borrowing at date T (is − ∂T log P (t. T ) is the same for Ct Q(t. T ). where
T
Σi (t. −1 Zt = Bt Ct Dt . T ) = dCt = Ct
ρi (t) dWti + λt dt . T ) : g(t. where
−1 Xt = Bt P (t. t)) the dollar cash bond (equal to exp
t r ds) 0 s
the sterling zerocoupon bond market prices
∂ the forward rate of sterling borrowing at date T (is − ∂T log Q (t. T ) : f (t. T ) = − Ti (t.
The differentials of these processes are
n
dt f (t. T ). T ))
the volatility of f (t. T ). T ) dWti + β (t. τi (t.
˜ Ti (t. T ) is the same for Q(t.
i=1 n
dt g (t. and ˜ Ti (t.6. T ). T ) + ρi (t) . T ) : τ (t.
. T ) dWti + α (t. −1 Yt = Bt Ct Q (t. T ) : rt : Bt : Q(t. T ) dt. Y and Z respectively. T ) =
i=1 n
σi (t. u) du.5. T ) the dollar short rate (equal to f (t.
T ) = P (t. u) . u)) .
where ξ(t. T ) =
i=1
νt = λt − rt + ut −
Then there will be a martingale measure only if there is some choice of γ which makes all of X . u) − α (t. T ) and νt are deﬁned to be
n
ξ (t.152
CHAPTER 6.) A derivative X paid in dollars at date T will have value at time t
−1 Vt = Bt EQ BT XFt . η(t. T )
i=1 n
ut −
i=1
ρi (t) Ti (t. Then the SDEs ˜ ˜ ˜ Brownian motion W t t t t 0 of X . . The ﬁrst thing to do is to ﬁnd a change of measure under which Xt . T )
i=1 n
˜ Σi (t. T ) dWti + ˜ ˜ Ti (t. BIGGER MODELS
Our plan. u)) du dt
t T
dYt = Yt
i=1 n
νt +
(η (t. T ) (γi (t) − Σi (t. where W i = W i + t γi (s)ds.
dCt = Ct
i=1
˜ ρi (t) dWti + (rt − ut ) dt . ˜ Ti (t. T ). much as ever. T ) . T ) dt . u)) du dt
t
dZt = Zt
i=1
˜ ρi (t) dWti + νt dt . T ) dWti +
n
dt Q (t. (And uniqueness will follow if the volatility vectors of any n of the dollar tradable securities make an invertible matrix. is to follow the three steps to replication.
. .
i
η (t. . This happens if
n
α (t. ρi (t) γi (t) . . Yt . For any previsible nvector γ = (γi (t))n . T )) . T ) =
i=1 n
σi (t.
Then under this Q measure
n
dt P (t. u) γi (t) − Ti (t. T ) =
i=1
λt = r t − u t +
i=1
ρi (t) γi (t). T ) = Q (t. ˜ τi (t. we will be able to hedge.
As long as this measure Q is unique. W n ). and Zt are all martingales. T ) =
i=1 n
σi (t. T ) γi (t) − Ti (t. u) (γi (t) − Σi (t. there is a new measure Q and a Qi=1 ˜ = (W 1 . ˜ τi (t.
n
β (t. Y and Z with respect to Q are
n
dXt = Xt
i=1 n
˜ Σi (t. Y and Z driftless. u) − β (t. T ) dWti +
T
(ξ (t. T ) dWti + rt dt .
T ) = Q (t. the sterling bonds have SDE
dt Q (t.
• equivalent martingale measure (EMM). A market is said to be complete if any possible derivative claim can
be hedged by trading with a selfﬁnancing portfolio of securities. Suppose we have a market of securities and a numeraire cash bond under a measure P.6. Then the martingale representation theorem gives a hedge for the derivative.
• arbitragefree. An EMM is a measure Q equivalent to P. dQ Bt ˜ As Zt has the SDE dZt = Zt i ρi (t) dWti the difference in drifts between the Q£ ˜ ˜ Brownian motion W £ and the QBrownian motion W is just ρ. This is just a more precise name for what we call the martingale measure. the sterling investor will agree with the dollar investor on prices of future payoffs. As explained in section 6. And there is. it is worth carefully laying out some concepts we have already brushed up against.
• complete. An arbitrage opportunity would be a (selfﬁnancing) trading strategy which started with zero value and terminated at some deﬁnite date T with a positive value.4.
To the sterling investor. ARBITRAGEFREE COMPLETE MODELS
153
The sterling investor
The sterling investor is on the other side of the mirror. T )
i=1
˜ Ti (t. Before stating this canonical theorem. under which the bonddiscounted securities are all Q
martingales. A market is arbitragefree if there is no way of making riskless
proﬁts.
6. (Normalizing D0 = 1/C0 for convenience.
which is exactly the form that HJM leads us to expect. T ) dWi£ (t) + ut dt . The repeated recurrence of this program suggests that there might be a more general result underpinning it.6. This reﬂects that his numeraire is the sterling cash bond Dt rather than the dollar cash bond. He works with a different martingale measure Q£ . The RadonNikodym derivative of Q£ with respect to Q will be the ratio of the dollar worth of the sterling bond to the dollar numeraire.6 Arbitragefree complete models
Time and again we have seen the same basic techniques used to price and hedge derivatives. Firstly. the CMG theorem is used to make the discounted price processes into martingales under a new measure. That is EQ ˜ ˜ Wi£ (t) = Wti −
n t 0
ρi (s)ds. A market is arbitragefree if there are absolutely no such arbitrage opportunities.) That is
Ct Dt dQ£ Ft = = Zt .
.
Arbitragefree and completeness theorem (Harrison and Pliska) Suppose we have a market of securities and a numeraire bond. We have not found an arbitrage opportunity. This ensures that there is an EMM Q.
In other words. The other key HJM condition is that the volatility matrix
Σi (t. T ) =
i=1
σi (t. but neither are we sure that one might not exist.
Martingales mean no arbitrage
A martingale is really the essence of a lack of arbitrage.
for some previsible processes γi (t). its future expectation. which means there is only one viable price of risk in the market. Tj )
n i. In both the binomial tree and BlackScholes models we found there was one and only one EMM. and (2) in which case. BIGGER MODELS
Already we have examples of the binomial trees and the continuoustime BlackScholes model. and consequently for the market to be complete. and γ is the market price of risk. We now see that this is to make sure that the model is arbitragefree. Although the technical details and exact deﬁnitions are passed over. Even more so in the multiple stock models (section 6. T ) (γi (t) − Σi (t. the structure of the following can be proved rigorously. and we were able to hedge claims. is just its current value at time s. and for all t less than T1 . Then (1) the market is arbitragefree if and only if there is at least one EMM Q. the market is complete if and only if there is exactly one such EMM Q and no other. The martingale is not ‘expected’ to be either higher or lower than its
. given the history up to time s. these conditions were also visible.154
CHAPTER 6. The governing rule for a Qmartingale Mt is that
EQ (Mt Fs ) = Ms . Both of these are complete markets with an EMM. The model demands that the forward rate drift α(t. In the HJM bondmarket model. This simple yet powerful theorem makes sense of our experience. There we could ﬁnd a market price of risk γt but it (and so Q too) was only unique if the volatility matrix Σt was invertible. T )).j=1
is nonsingular for all sequences of dates T1 < · · · < Tn . And it was exactly that invertibility which lets us hedge. It is worth getting a feel of why this theorem works. This is sufﬁcient (but actually slightly more than necessary) for the EMM to be unique. T ) satisﬁed
n
α (t.3).
An arbitrage opportunity. if an EMM exists. ARBITRAGEFREE COMPLETE MODELS
155
present value.6. the digitallike claim which pays off the cash bond value at time T if A has happened has payoff X = BT IA .
and it satisﬁes the selfﬁnancing equation
dVt = φt dSt + ψt dBt . so it must be hedgeable. Or in our language. (We assumed that we could hedge all claims.
−1 We can calculate the discounted value of the portfolio Et = Bt Vt .) This is a valid derivative.
. Any strategy can make no more than nothing from nothing.
Hedging means unique prices
If we can hedge. and then
dEt = φt dZt . suppose that we could hedge. But the Qexpectation of ET is zero. A martingale is essentially a ‘fair game’ and any strategy which involves only playing fair games cannot guarantee a proﬁt. To see this. Et is a Qmartingale because Zt is. From which it is clear that VT is zero as well.) So there must be a selfﬁnancing portfolio (φ. then there can only be at most one EMM. Can this really be an arbitrage opportunity? Crucially.6. there are no arbitrage opportunities. ψ). (Assuming for simplicity a two security market of stock St and bond Bt . but that there are two different EMMs Q and Q . For any event A in the history FT . on the other hand. ψ) which hedges X .) Then its value at time t is
Vt = φt St + ψt Bt . is a oneway bet which is certain to end up higher than it started. (The indicator function IA takes the value 1 if the event A happens.
−1 As usual the discounted claim Et = Bt Vt satisﬁes
dEt = φt dZt . with value
Vt = φt St + ψt Bt . Suppose now that the strategy does start with zero value (V0 = 0) and ﬁnishes with a nonnegative payoff (VT ≥ 0). and zero otherwise.
−1 where Zt is the discounted stock price Bt St which is a Qmartingale. so the only possible value that ET can take is zero too. Suppose we have a potential arbitrage opportunity contained in the selfﬁnancing portfolio strategy (φ.
−1 But VT ≥ 0 and (because BT > 0) so is ET ≥ 0. And so
EQ (ET ) = EQ (ET F0 ) = E0 = V0 = 0.
and so E0 = Q(A) = Q (A). there has been other work. So also must Et be.156
CHAPTER 6. and thus Q = Q . Now Zt is both a Qmartingale and a Q martingale as both Q and Q are EMMs.
But ET is just the indicator function of the event A. We showed that if there was an EMM there was no arbitrage. Also we proved that hedging can only happen with a unique EMM. For the continuous case and more advanced models.
Harrison and Pliska
We have only proved each result in one direction. but did not show that if there is no arbitrage then there actually is an EMM. But the increasing technicality of this should not stand in the way of an appreciation of the remarkable insight of Harrison and Pliska. And from that. in Stochastic Processes and their Applications (see appendix A for more details). If any two EMMs are identical. we see
E0 = EQ (ET ) = EQ (ET ) . IA . The full and rigorous proofs of all these results in the discretetime case are in the paper ‘Martingales and stochastic integrals in the theory of continuous trading’ by Michael Harrison and Stanley Pliska.
. notably by Delbaen and Schachermayer. BIGGER MODELS
−1 where Zt is the discounted stock price Bt St . but not that the uniqueness of the EMM forced hedging to be possible. the two measures agree completely. As A was completely general. The two measures Q and Q which were trying to be different actually give the same likelihood for the event A. then there can only really be one EMM.
Springer (2nd edition 1994. Both Revuz and Yor. 550 pages)
• Diffusions. David Williams. and the representation o theorem. Ross is an introduction to the basic (static) probabilistic ideas of events. CameronMartinGirsanov change of measure. distribution and expectation.
Probability and stochastic calculus books
• A ﬁrst course in probability. The lists below have been kept short. in the hope that in this case less choice is more.Appendix A
Further reading
The longer a list of books is. Chris Rogers o
and David Williams. Grimmett and Stirzaker contain that material in their ﬁrst half. Cambridge University Press (1991. and Rogers and Williams provide a detailed technical coverage of stochastic calculus. but also is an excellent introduction to martingales themselves. 475 pages) These books are arranged in increasing degrees of technicality and depth (with the last two being at an equivalent level) and contain the probabilistic material used in chapters one. Sheldon Ross. 540 pages)
• Probability with martingales. Wiley (1987. two and three.
250 pages)
• Continuous martingales and Brownian motion.
157
. as well as the development of random processes including some basic material on martingales and Brownian motion. Daniel Revuz and Mark Yor. a reader with background knowledge will ﬁnd them invaluable and deﬁnitive on questions of stochastic analysis. and martingales: vol. the fewer will actually be referred to. Markov processes. (conditional) expectation and measures. They both contain all our tools. Probability with martingales not only lays the groundwork for integration. Geoffrey Grimmett and David Stirzaker. Although dense with material. Ox
ford University Press (2nd edition 1992. stochastic differentials. 2 Itˆ calculus. Macmillan (4th edition 1994. Itˆ ’s formula. There is also a chapter containing a simple representation theorem and a discretetime version of BlackScholes. 420
pages)
• Probability and random processes. likelihood.
231–237. but still accessible. and other derivative securities. 141–183. 450 pages) Hull is a popular book with practitioners.
Chapter four: pricing market securities
Some notable journal papers include:
• The pricing of options and corporate liabilities. R C Merton. The chapterbychapter bibliographies are another useful feature. many pricing problems become differential equation problems. but it is still fascinating to see how the subject began. Paul Wilmott. PrenticeHall (1985. M B Garman and S W Kohlhagen. 490 pages)
• Dynamic asset pricing theory.
• Theory of rational option pricing. Merton provides a more rigorous treatment. 500 pages). p.
• Foreign currency option values. 4 (1973). J C Cox and M Rubinstein. Eventually. Oxford Financial Press’s volume comes at the subject purely from a differential equation framework without using stochastic techniques. Garman and
. Oxford Financial Press (1993. John Hull. 2 (1983). 637–654.
300 pages)
• Option pricing: mathematical models and computation. and makes extension to dividendpaying stocks and a barrier option. Journal of
International Money and Finance. Journal
of Political Economy. For readers with mathematical backgrounds. Dufﬁe is a much more mathematically rigorous text. it is a good read.
• Options markets. Bell Journal of Economics and
Management Science. but unless a reader has experience in this area. F Black and M Scholes. and numerical procedures for implementation are also included. 81 (1973). At the time they were as concerned with pricing the stock of companies with outstanding liabilities (such as corporate bonds or warrants) as they were about options and derivatives. M Rubinstein. Darrell Dufﬁe. not the technical detail. contemporaneously with BlackScholes. He contains sections on equilibrium pricing and optimal portfolio selection as well as a treatment of continuoustime arbitragefree pricing along the same lines as this book. (May 1991). though the paper should be read for its insights. PrenticeHall (2nd
edition 1993. • Two into one. Jeff Dewynne
and Sam Howison.158
APPENDIX A. it is not necessarily the best place to start from.futures. RISK. Princeton University Press (1992. FURTHER READING
Financial books
• Options. A number of models are discussed. 49. laying out the various realworld options contracts and markets before starting his analysis.
The BlackScholes paper is now of historical interest.
159
Kohlhagen described foreign exchange options, whilst Cox and Rubinstein contain some exotic option formulas, amongst much else. The Rubinstein paper from RISK is concerned with quantos and crosscurrency options.
Chapter ﬁve: interest rates
In the interestrate setting, HeathJarrowMorton is as seminal as Black Scholes. By focusing on forward rates and especially by giving a careful stochastic treatment, they produced the most general (ﬁnite) Brownian interestrate model possible. Other models may claim differently, but they are just HJM with different notation. The paper itself repays reading and rereading.
• Bond pricing and the term structure of interest rates: a new methodology for
contingent claims valuation, David Heath, Robert Jarrow and Andrew Morton, Econometrica, 60 (1992), 77–105. In addition to the HJM paper, notable papers on the various interestrate market models include
• Term structure movements and pricing interest rate contingent claims, T S Y Ho
and SB Lee, Journal of Finance, 41 (1986), 101 1–1029.
• An equilibrium characterization of the term structure, O A Vasicek, Journal of
Finance, 5 (1977), 177–188.
• Pricing interest rate derivative securities, J Hull and A White, The Review of
Financial Studies, 3 (1990), 573–592.
• A theory of the term structure of interest rates, J C Cox, J E Ingersoll and S A
Ross, Econometrica, 53 (1985), 385–407.
• Bond and option pricing when short rates are lognormal, F Black and P Karasin
ski, Financial Analysts Journal, (JulyAugust 1991), 52–59.
• The market model of interest rate dynamics, A Brace, D Gatarek and M Musiela,
UNSW Preprint, Department of Statistics S95–2.
• Which model for the termstructure of interest rates should one use? L C G
Rogers, in Mathematical Finance (ed. M H A Davis, D Dufﬁe, et al.), IMA Volume 65, SpringerVerlag, 93–116. The last of these is a review paper of models and their properties, whilst the others describe separately all the major models considered in the chapter.
Chapter six: bigger models
• Martingales and stochastic integrals in the theory of continuous trading, Michael
Harrison and Stanley Pliska, Stochastic Processes and their Applications, 11 (1981), 215–260.
160
APPENDIX A. FURTHER READING
• The fundamental theorem of asset pricing, F Delbaen and W Schachermayer,
Mathematische Annalen, 300 (1994), 463–520.
• The valuation of options for alternative stochastic processes, J C Cox and S A
Ross, Journal of Financial Economics, 3 (1976), 145–166. Harrison and Pliska made the next step forward by linking, in a general framework, the absence of arbitrage to the existence of a martingale measure, and showing that the ability to hedge depended on there only being one such measure. That this idea still underpins much of ﬁnancial mathematics today is a demonstration of the importance of the paper. Delbaen and Schachermayer go over similar ground but in a much more technical way to deal with the particular problems of continuoustime processes, including discontinuous processes. Cox and Ross cover option pricing for models more general than BlackScholes, including those paying dividends.
Appendix B
Notation
Notation can be divided naturally into three parts: lower case (generally deterministic), upper case (generally random), and Greek.
Lower case
a (real) parameter c a constant; coupon rate dQ RadonNikodym derivative of Q with respect to P dP dt inﬁnitesimal time increment dWt inﬁnitesimal Brownian increment f a function fP (x) probability density function of the law P f (t, T ) bond forward rates g a function g(x, t, T ) the function (− log P (t, T )rt = x) i an integer j an integer k contract strike/exercise price; an integer; an offset n an integer n[t] number of dividend payments made by time t p, pj a probability q, qj a probability r constant interest rate rt variable interest rate process; instantaneous rate s initial stock price, alternative time variable sj possible value for the discrete stock process t time u foreign currency interest rate; real variable
a
161
} a normal random variable with mean µ and variance σ 2 hypothetical discrete derivative price a probability measure forward measure bond prices a probability measure the ndimensional real vector space bond yield surface
. . T ) FQ Fi Ft IA I(t) K L(T ) L(t. a constant HJM volatility matrix bond price process solution of a Riccati equation foreign exchange rate. T ) E EP Et F Fs (t. 2. coupon bond price.162
APPENDIX B. T ) Q Rn R(t. T )
an event. Bt B(t. T ) quanto forward price history of discrete stockprice process up to ticktime i history of Brownian motion up to time t indicator function of the event A sequence number of next coupon payment option strike price LIBOR rate forward LIBOR rate a martingale a martingale the set of nonnegative integers {0. . horizontal axis variable timedependent factor of volatility surface maturitydependent factor of volatility surface
Upper case
A At Bi . T ) Mt Nt N N (µ. T ) Ct Di Dt D(t. NOTATION
x xi (t) yi (T )
a real variable. σ 2 ) P P PT P (t. . numeraire ﬁnancing gap foreign currency cash bond solution of a Riccati equation expectation operator expectation under the measure P discounted portfolio value process forward price forward price at time s for P (t. 1.
S t
T Ti Ut V Vt V (s. . T ) γt γi (t. T ) ˜ Zt
stock price process tradable asset price stock price processes maturity/exercise time of a derivative coupon payment times foreign currency derivative value process derivative value derivative value process BlackScholes option price random walk Brownian motion Brownian motion independent Brownian motions random variable. claim value of a derivative sequence of random variables a stochastic process a stochastic process integral of yi over [t. ∆Vi ζt θ θt λ
a real parameter forward rate drift a function of two variables (Vasicek model) change of measure drift.163
Si . T ) Wn (t) Wt ˜ Wt Wt1 . T ) δ δt δSi . market price of risk BGM volatility surface dividend yield. coupon payment interval a small time increment branch widths change in value across δt of Si . δni ∆Si . . T ) Z Zi . T ) β(t. . Vi . etc change of measure process a real variable deterministic drift function a real parameter
. S t ˜ St
1 n St . . Wtn X Xi Xt Yt Yi (t. . . T ] a (normal) random variable discounted stockprice process discounted bond prices discounted tradable asset price
Greek case
α α(t. Z t Z(t.
NOTATION
µ µt νt πi Πi ρ ρ ¯ ρt σ σ1 . T ) τ ˜ φt .164
APPENDIX B. stopping time stockholding strategy. representation theorem integrand normal distribution function: Φ(x) = P N (0. maturity date. 1) ≤ x bondholding trading strategy a sample path
. φt Φ ψt ω
constant stock drift variable stock drift process stock drift process path probability portfolio correlation the orthogonal complement 1 − ρ2 volatility process constant stock volatility stock volatilities variable stock volatility process forward rate volatility surface multifactor forward rate volatility surface term volatility volatility matrix bond price volatilities time horizon. T ) σ ¯ Σt Σ(t. T ) σi (t. σ2 σt σ(t.
each of whose nodes branches into two at the next stage BlackScholes a stock market model with an analytic option pricing formula Bonds interest bearing securities which can either make regular interest payments and/or a lump sum payment at maturity Bond options an option to buy or sell a bond at a future date Brownian motion the basic stochastic process formed by taking the limit of ﬁner and ﬁner random walks.Appendix C
Glossary of technical terms
Adapted a process which depends only on the current position and past movements of the driving processes. that it is meanreverting Average the arithmetic mean of a sample Bank account process an account which is continuously compounded at the prevailing instantaneous rate. and behaves like the cash bond Binomial process a process on a binomial tree Binomial representa.a discretetime version of the martingale representation tion theorem theorem on the binomial tree Binomial tree a tree. It is a martingale. It is unable to see into the future American call option a call option which can be exercised at any time up to the option expiry date Arbitrage the making of a guaranteed riskfree proﬁt with a trade or series of trades in the market Arbitrage free a market which has no opportunities for riskfree proﬁt Arbitrage price the only price for a security that allows no arbitrage opportunity Autoregressive of a process. and is not Newtonian differentiable
165
. with zero drift and unit volatility.
only if this difference is positive. such as P(X ≤ xF) Conditional expecta. for conditioning on the history of the process up to time t Contingent claim a claim whose amount is determined by the behavior of market securities up until the time it is paid
Calculus
. such as gold.166
APPENDIX C. which says that the average of a sample of IID random variables is asymptotically normally distributed Change of measure viewing the same stochastic process under a different set of likelihoods. but not Brownian motion which requires the techniques of stochastic calculus. is two. given that the ﬁrst toss was heads. [From calculus (Lat. oil or frozen concentrated orange juice Complete market a market in which every claim is hedgable Conditional distribu.taking an expectation given some history as known. a pebble used in an abacus] Call option the option to buy a security at/by a future date for a price speciﬁed now CameronMartina result which interprets equivalent change of measure as Girsanov theorem changing the drift of a Brownian motion Cap a contract which periodically pays the difference between current interest rate returns and a rate speciﬁed at the start. For tion instance the conational expectation of the number of heads obtained in three tosses. GLOSSARY OF TECHNICAL TERMS
generally a formal system of calculation. Newtonian calculus handles smooth functions. A cap can be used to protect a borrower against ﬂoating interest rates being too high Caplet an individual cap payment at some instant Cash bond a liquid continuously compounded bond which appreciates at the instantaneous interest rate Central Limit theorem a statistical result. changing the probabilities of various events occurring Claim a payment which will be made in the future according to a contract Commodity a real thing. whereas the unconditioned expectation is only one and a half.the distribution of a random variable conditional on some tion information F . Written E(·Ft ).). in particular concerned with analyzing behavior in terms of inﬁnitesimal changes of the variables.
f (x)dx is the probability that X lies in the interval [x. If one variable gets larger as the other does. as
E(X ) =
−∞ 2 ∞
Continuous
x2 f (x)dx
Derivative
a security whose value is dependent on (derived from) existing underlying market securities. the correlation is positive. x + dx]. and so forth. Formally correlation is the covariance of the random variables divided by the square root of the product of their individual variances Coupon a periodic payment made by a bond Covariance a measure of the relationship of two random variables. leading to exponential growth Contract an agreement under law between two principals. the covariance is zero if the variables are independent (and vice versa in the case of jointly normal random variables). rather than annually or pounded monthly. allowing inﬁnite divisibility of time Continuously com. Intuitively.interest is compounded instantly. The function f is nonnegative. or counterparties Correlation a measure of the linear dependence of two random variables.see normal distribution function tegral Currency the monetary unit of a country or group of countries Default free there being no chance that the bond issuer will be unable to meet his ﬁnancial undertakings (used theoretically) Density the probability density function f is the derivative (if it exists) of the distribution function of a continuous random variable. and negative if one gets larger as the other gets smaller. and can be used to calculate expectations. Formally the covariance of two variables is the expectation of their product less the product of their expectations Cumulative normal in. integrates to one. See also contingent claim
. The limits of one and minus one correspond to exact dependence. whereas independent variables have zero correlation.167
a process or function which only changes by a small amount when its variable or parameter is altered inﬁnitesimally Continuoustime a process which depends on a realvalued time parameter.
168
APPENDIX C. the description of the likelihood of its every possible value the (cumulative) distribution function F of a random variable is deﬁned so that F (x) is the probability that the random variable is no larger than x.}
Dividends Dol´ ans exponential e
of a random variable. . Compare with American call option a set future date at which an option may be exercised or not see strike price
. and nothing otherwise scaling a future reward or cost down to reﬂect the importance of now over later a bond which promises to make a lump sum payment at a future date. such as from the sets N or
{0. . If F is differentiable. then its derivative is the density regular but variable payments made by an equity for a local martingale Mt . to ﬁnd the sequence (xn ) which obeys
axn+2 + bxn+1 + cxn = d
Diffusion Digital Discount Discount bond Discrete Distribution Distribution function
a stochastic process which is the solution to a SDE a derivative which pays off a ﬁxed amount if a given future event happens. 2δt. The function F increases (weakly) from 0 to 1. separated values. which is another local martingale Xt =
exp Mt −
1 t 2 2 0 (dMs )
Drift Driftless Equilibrium distribution Equities Equivalent martingale measure (EMM) Equivalent measures European call option Exercise date Exercise price
the coefﬁcient of the dt term of a stochastic process a process with constant zero drift a distribution of a process which is stable under time evolution stocks which make dividend payments see martingale measure two measures P and Q are equivalent if they agree on which events have zero probability a call option which can be exercised or not only at the option exercise date. For example. but until then is worth less than its face value taking distinct. . GLOSSARY OF TECHNICAL TERMS
Difference equation
the discrete analogue of a differential equation. δt. this is the solution of the SDE dXt = Xt dMt .
which itself is a e gales (local) martingale Filtration the history. A ﬂoor can be used to protect a lender against ﬂoating interest rates being too low. and all of whose joint distributions are jointly normal HeathJarrowMorton a model of the interestrate market (HJM)
. of a process.5 Future a forward traded on an exchange FX abbreviation for foreign exchange Gaussian process a process. A straight line is a fractal of dimension one. which will quickly either become standard products or will sink without trace the mean of a random variable.the Dol´ ans exponential of a martingale. and a Brownian motion path is a fractal of dimension 1.169
Exotics Expectation
new derivative securities. (Ft )t≥0 . which will be the limiting value of the average of an inﬁnite number of identical trials. where Ft is the information about the path of the process up to time t Fixed of interest rates. all of whose marginals are normally distributed. only smaller. that they can move with the market over the term of the contract Floor a contract which periodically pays the difference between a rate speciﬁed at the start and current interest rate returns.
E(X) =
−∞
xf (x)dx
Exponential Brownian a process which is the exponential of a drifting Brownian motion motion Exponential martin. that they are constant throughout the term of the contract Floating of interest rates. See also cap Floorlet which is to ﬂoors as caplets are to caps Foreign exchange the market which prices one currency in terms of another Forward an agreement to buy or sell something at a future date for a set price. only if this difference is positive. called the forward price Forward rate the forward price of instantaneous borrowing Fractal a geometrical shape which on a smallscale looks the same as the largescale. For a discrete and a continuous random variable (with density f ) it is respectively
∞ ∞
E(X) =
n=0
nP(X = n).
but not necessarily a martingale Logdrift of a stochastic process Xt .170
APPENDIX C. which itself then implies the next case.a random variable whose logarithm is normally distributed tion Logvolatility of Xt is the volatility of log Xt . If Xt has volatility σt and drift µt . A daily set of interest rates for various currencies and maturities Local martingale a stochastic process which is driftless. the drift of log Xt Lognormal distribu. and so on the rate of interest paid on a very very short term loan tradable securities or contracts the rate at which interest is paid the market which determines the time value of money a stochastic version of the ‘chain rule’ which expresses the volatility and drift of the function of a stochastic process in terms of the volatility and drift of the process itself and the derivatives of the function. then Yt = f (Xt ) has volatility f (Xt )σt and drift
2 f (Xt )µt + 1 f (t)σt 2
Kolmogorov’s strong see strong law law Law of the unconscious the result that if a random variable X has density f . involving the demonstration that the current case follows from the previous case. GLOSSARY OF TECHNICAL TERMS
Hedge History Identically distributed IID Independent Indicator function Induction
Instantaneous rate Instruments Interest rate Interest rate market Itˆ ’s formula o
to protect a position against the risk of market movements the information recording the path of a process of random variables. none of which have any relation or inﬂuence on any of the others a function of a set which is one when the argument lies in the set and zero when it is outside a method of proof. have the same probabilistic distribution abbreviation for Independent. or equivalently the volatility of dXt /Xt Long (of position) having a positive holding
. then the statistician expectation of h(X) is
∞
E h(X) =
−∞
h(x)f (x)dx
LIBOR
the London InterBank Offer Rate. Identically Distributed of variables.
is its current value. describing how likely each one is Multifactor a market model which i s driven by more than one Brownian motion Newtonian calculus classical differential and integral calculus. conditional on the present Martingale a process whose expected future value. conditional on the past. E(Mt Fs ) equals Ms for every s less than t Martingale measure a measure under which a process is a martingale Martingale representa. Two processes may be different. written N (µ. Commonly situated in electronic space Market maker (in UK) a dealer who is obligated to quote and trade at twoway prices Market price of risk a standardized reward from risky investments in terms of extra growth rate Markov of a process. or more generally the time at which any claim pays off Mean synonym for expectation Mean reversion the property of a process which ensures that it keeps returning to its longterm average Measure a collection of probabilities on the set of all possible outcomes. That is.171
Marginal
the marginal distribution of a process X at time t is the distribution of Xt considered as a random variable in isolation.a result which allows one martingale to be written as the tion theorem integral of a previsible process with respect to another martingale Maturity the time at which a bond will repay its principal. meaning that its future behavior is independent of its past. yet have exactly the same marginal distributions Market a place for the exchanging of price information. parameterized by a mean µ and variance σ 2 . relating to smooth or differentiable functions Newtonian function a function which is smooth enough to have a classical (Newtonian) derivative Node a point on a tree where branches start and ﬁnish Noise a loose term for volatility Normal distribution a continuous distribution. σ 2 ) with density
f (x) = √ 1 2πσ 2 exp − (x − µ)2 2σ 2
.
The probability will be the product of the probabilities of the individual branches taken a payment abbreviation for Partial Differential Equation a type of random process with discontinuities a collection of security holdings the amount of a security held. without the mediation of an exchange the probability of a tree process taking a particular path through the tree. which can either be positive (a long position) or negative (a short position) a stochastic process which is adapted and is either continuous or leftcontinuous with rightlimits or is a limit of such processes the face value that a bond will pay back at maturity the chance of an event occurring a sequence of random variables.172
APPENDIX C. derivatives which pay off in another currency of one measure with respect to another is the relative likelihood of each sample path under one measure compared with the other a function of a sample space
. GLOSSARY OF TECHNICAL TERMS
Normal distribution the distribution function of the normal random variable. Often the cash bond ODE abbreviation for Ordinary Differential Equation Option a contract which gives the right but not the obligation to do something at a future date OrnsteinUhlenbeck a mean reverting stochastic process with SDE (OU) process
dXt = σdWt + (θ − αXt )dt
Overthecounter Path probability
Payoff PDE Poisson process Portfolio Position Previsible
Principal Probability Process Product rule Putcall parity Quantos RadonNikodym derivative Random variable
an agreement concluded directly between two parties. function written Φ(x) = P(N (0. 1) ≤ X) Numeraire a basic security relative to which the value of other securities can be judged. parameterized by time a result giving the stochastic differential of the product of two stochastic processes the observation that the worth of a call less the price of a put struck at the same price is the current worth of a forward crosscurrency contracts.
A simple symmetric random walk is Nvalued and after each time step goes up one with probability 1 and down one with probability 1 2 2 a tree where branches can come together again a selfﬁnancing portfolio trading strategy which hedges a claim precisely no chance of anything going wrong a martingale measure abbreviation for Stochastic Differential Equation a piece of paper representing a promise a strategy which never needs to be topped up with extra cash nor can ever afford withdrawals a process which can be decomposed into a local martingale term and a drift term of ﬁnite variation (in UK) a stock or equity (of position) having a negative. or vice versa an option to enter into a swap agreement at a future date
. or borrowed. given some technical conditions an agreement to make a series of ﬁxed payments over time and receive a corresponding series of payments dependent on current interest rates. such as those involving Brownian motion terms a continuous process. holding see instantaneous rate a market model which is driven by only one Brownian motion the square root of the variance synonym for random a calculus for random processes.173
Random walk
Recombinant tree Replicating strategy Risk free Riskneutral measure SDE Security Selfﬁnancing Semimartingale Share Short Short rate Singlefactor Standard deviation Stochastic Stochastic calculus Stochastic process Stock Stock market Strike price Strong law
Swaps
Swaption
a discrete Markov process made up of the sum of a number of independent steps. which can be decomposed into a Brownian motion term and a drift term a security representing partial ownership of a company a place for trading stocks the price at which an asset may be bought or sold under an option the result that the average of a sample of n IID random variables will converge to the mean of the distribution as n increases.
that it can be traded either directly. the coefﬁcient of the Brownian motion term of a stochastic process the result that the average of n IID random variables is increasingly less likely to be signiﬁcantly different from the distribution mean as n increases
. a choice which may depend on market movements a charge for buying or selling a security a graph of nodes linked by branches which contains no closed loops or circuits a basic market security. and the length of the loans the variance of the logarithm of a security price over a time period. the expectation of its square less the square of its expectation. its square is the term variance divided by the length of the term:
σ 2 = Var log(ST /S0 ) /T ¯
Time value of money Tower law Tradable Trading strategy Transaction cost Tree Underlying Vanilla Variable coupon Variance
Volatility
Weak law
the difference between cash now. bonds and currencies of a product. that is
f (x + h) = f (x) + hf (x) + 1 h2 f (x) + 1 f (x) · · · 2 6
Term structure Term variance Term volatility
the relationship between the interest rates demanded on loans. for s < t of an asset. or equivalently the expected square of the difference between the random variable and its mean the amount of ‘noise’ or variability of a process. GLOSSARY OF TECHNICAL TERMS
Taylor expansion
for Newtonian functions. Explicitly. or indirectly by trading a matching portfolio a continuous choice of portfolio. more precisely. Var log(ST /S0 ) the effective (annualized) volatility of an asset over a time period. Formally. the expression of the value of a function f near x in terms of the value of it and its derivatives at x. and cash later which is subject to a discount the result that E E(XFt )Fs = E(XFs ).174
APPENDIX C. such as stocks. the standard basic version periodic payments from a ﬂoating interestrate contract a measure of the uncertainty of a random variable.
175
Wiener process With probability 1
Yield Yield curve Zero coupon
synonym for Brownian motion of an event. for example. having probability one of occurring. This is not quite the same as being guaranteed for sure. but with probability one it will not the average interest rate offered by a bond the graph of yield plotted against bond maturity a bond which does not make any payments until maturity
. a normal random variable can take the value zero. as.