Sie sind auf Seite 1von 139

SOR 3430

The Mathematics of Financial


Markets: Discrete Models
Organization
Subject
4 ECTS
1 hour per week for 28 hours + tutorials
Course held during semesters 1 and 2
Method of assessment: Examination

Lecturer: Ms. Natalie Attard


Email: natalie.attard@um.edu.mt
Room: 513

Contents
An Introduction To Financial Markets
Single Period Financial Markets
The Optimal Portfolio and Optimal Consumption-Investment Problems in
Single Period Financial Markets
Multiperiod Financial Markets
The Optimal Portfolio and Optimal Consumption-Investment Problems in
Multiperiod Financial Markets

Suggested Text
1.

Pliska, S., (1997) Introduction to Mathematical Finance : Discrete Time


Models, Blackwell

2.

Hull, J C. (1993) Options, Futures, and Other Derivative Securities, Prentice


Hall Inc.

3.

Galitz , L. (1995) Financial Engineering, Pitman Publishing

4.

Wilmott, P., Howison, S. and Dewynne, J. (1999) The Mathematics of


Financial Derivatives, Cambridge University

5.

Elliott, R.J., and Kopp, P.E., (1999) Mathematics of Financial Markets,


Springer

6.

Capinski, M., and Zastawniak, T., (2003) Mathematics for Finance: An


Introduction to Financial Engineering, Springer

7.

Cvitani, J., and Zapatero, F., (2004) Economics and Mathematics of


Financial Markets, MIT Press

Chapter 1
An introduction to Financial Markets
Governments, multinational and national corporations, international organizations,
firms and businesses, banks, financial institutions and individuals, from time to
time, need to raise money more than they have available at specific times to fund
their activities. Conversely, these same economic agents, from time to time, have
funds available to them which they do not need to make immediate use of and
would not want to keep inactive.

They would prefer to invest such funds.

Financial Markets offer opportunities for these actors to facilitate the lending and
borrowing of funds. In transactions involving the transfer of funds it is usual to
expect the borrower to be given access to funds provided that she pays them back
eventually plus some agreed percentage as interest over time. This extra (return)
covers the risk and compensation for the deprivation which the lender has to take
on.
Financial markets have existed for centuries. In the last 40 years they have suffered
tremendous changes which have brought along a degree of sophistication quite
unparalleled in any other phase of the history of finance. It may be stated without
exaggeration that the world of finance and insurance has become the most
aggressive employer of Ph.D. graduates with a deep knowledge of mathematics of
the last 20 years in most countries with developed economies.
For over 30 years there has been a veritable revolution, silent and in constant
evolution, within the world of global finance with the impetus coming from two

major sources. During the 70s there was a switch from fixed exchange rates to
floating rates for currencies in developed countries. This forced a lot of important
currencies to revalue themselves radically in response to the real strength of the
economies which supported them.
The second most important contributor to radical changes in the operations of the
financial world has been the result of the successful convergence of
telecommunications and computer technology. This has meant in everyday practice
global dissemination of price-sensitive information and rapid response to it. Not
only is it very easy and inexpensive to get financial information, it is also very easy
to make deals. The dizzying speed with which the world of finance plunged into an
unknown future when it changed to electronic gear is held responsible for the great
crash of 1987.
Another important, more localized, source of change in nature is the tendency of
major exchanges to merge, regroup and make collaborative agreements amongst
themselves. The introduction of the Euro in 1999 was intended along this line.
Such attempts at synergies have resulted into a trend towards expanding services.
Servicing a global economy 24 hours a day needs collaborative effort to reduce
operating and trading costs.
The result of many of these changes has been an enormously increasing bulk of
business and an incredible increase in the velocity with which transactions are
effected. This in turn has meant a large increase in volatility of market rates.
Uncertainty in markets increased as opportunities and risks multiplied. There were
large pickings promised to those engaged in popular and flourishing academic
studies centred on the mathematical modelling of various mechanisms of financial
4

markets. The intrinsically dynamic aspect of the underlying situations, coupled


with inherently probabilistic underpinnings made time series and stochastic
processes the key mathematical branches to refer to.
The ongoing financial crises is the current major contributor to the radical changes
in the operation of financial markets and this is making room for debate about the
role of current mathematical models and the mathematicians employed in the
financial industry.

However, despite the much publicised catastrophe in financial

markets there is little evidence that financial mathematics have become less
important. Instead, the need for more sophisticated mathematics models and
understanding of financial markets has increased.
Before we plunge into mathematical models, theorems and formulas we first study a
bit the structure of capital markets and the way they operate.

1.1

Capital Markets

Financial markets can be described as mechanisms enabling the trading of money


and funds through the exchange of financial instruments between lenders and
borrowers. The financial needs of modern day economies are complex and varied
in nature. So there exist many specialized markets which have been designed and
developed to satisfy very special needs. First we mention the capital markets,
which are usually categorized as:
o Bond markets
o Equities markets
The main purpose of capital markets is to raise capital as need arises within
companies, countries and financial institutions. Such markets bring together
5

entities looking for funds while paying for their use, and investors willing to invest
money expecting to obtain returns on their investments. These markets, in contrast
with money markets (refer to section 1.2), cater for long term investments.

1.1.1

Bond Markets

Suppose an investor deposits money in a savings account say. The investor is


actually lending money to that bank, so the bank then pays back the investor
interest on this deposit. In a similar fashion, investors who invest in bonds make
loans to the issuer of the bond (i.e. a corporation or the government). This makes
the investors creditors (i.e. a person to whom money is owed by an institution, but
is not an owner of that institution). The issuer then pays a predetermined amount
of interest to the investor on a regular basis until the maturity date. Formally, a
bond is a contract in which authorized issuers owe holders an obligation to
repay an amount of money, called the principal which holders lend to issuers,
as well as interest (the coupon) at a later date (maturity).
Virtually bonds are issued for maturities greater than 1 year, most are in the range
1 to 10 years, and some may even be perpetual with no fixed maturity date.
Government agencies and large corporations are the major issuers of bonds. The
main types of bonds are:
o Government bonds (bonds issued by the government in the currency of the
respective country)
o Corporate bonds (bonds issued by a corporation a list of the current
corporate bonds listed on the Malta Stock Exchange can be found here:
http://online.hsbc.com.mt/pdf/corpbondprices.pdf)
o Floating-rate notes(FRN) (bonds with a variable interest rate, which are
hence subject to the interest rate risk)

o Eurobonds (international bonds which are not issued in the currency of the
country of issue)
o Medium-term-notes (MTN) (a debt security with maturity date ranging from
5 to 10 years).
A major difference between bonds and stocks is that holders of stocks are also
partial owners of the issuing company, whereas bond-holders are simply lenders.

1.1.2

Equities Markets

An equity or common share is a participation in the ownership of a company.


A share has a face value which is purely nominal with no promise to repay
that face value at any time. A company issuing shares has sole discretion each
year to the size of the dividend, if it is to be given at all.

On liquidation

(dissolution of the company or part of the company, and so its assets are
distributed) shareholders will be the last to receive any benefit from sale of assets.
Equities thus carry a lot of risk but they can provide better returns, and over the
long term they outperform debt securities. Determining the right price for such
assets is a major problem in the mathematics of finance. One approach, the
Capital Asset Pricing Model (CAPM) seeks to relate the risk and return
available from different investments. The New York Stock Exhange (NYSE)
and the National Association of Securities Dealers Automated Quotations
(NASDAQ) are the largest equity markets with the London and Tokyo Stock
Exchange following.

1.2 Money Markets

Money markets are characterized by borrowing and lending large amounts of


money for a short period of time. Maturities usually range from overnight to one
year. The typical size of deal will range from 200k to 40m. Some of this dept is
negotiable and can be bought and sold on a secondary market (i.e. a financial
market where previously issued financial instruments are bought and sold. In
contrast with primary markets, secondary markets do not issue new securities. In
Malta for example, only secondary markets exist, due to the small size of the
market) or it can be non-negotiable involving transactions between counterparties
who do not trade this dept in markets.
The negotiable debt consists of all public debt that is either traded on an exchange
or another market. Its trade is usually based on negotiable paper and traded
instruments. These may be defined as unconditional written orders to repay a debt,
easily transferable from one person to another. They include:
o Treasury bills (issued by the treasury department of a national government)
o Trade bills (issued by a commercial company)
o Bankers acceptances
o Bank certificates of deposit
o Commercial paper
o Euro-commercial paper
o Euronotes
For most of these instruments, secondary markets exist, allowing seller and buyers
to trade debt prior to maturity. The price at which an instrument is traded or issued
depends on time to maturity, credit quality, current interest rates, interest accrued
and other factors.

Non-negotiable debt, as its name implies, cannot be traded in open markets but is
contracted only in primary markets. It includes:
o Interbank deposits
o Federal funds(US)
o Repos and reverses
o Local authority and finance house deposits (UK)
Costs of borrowing can be computed, and compared across different sources,
through interest rates. These are quoted either:
o On a discount yield basis (i.e. as a percentage of face value converted to
annual basis)
o On a money market yield basis (as a percentage of the current price)
Furthermore quotations do not all use the same base: some 360-day years, others
365-day year.

1.3 Foreign Exchange


The foreign exchange is the international forum for exchange of currencies.
This type of market transaction occurs when two parties use different
currencies in an exchange of two amounts in the two different currencies.
The foreign exchange market is the most extensive and largest financial market
globally, involving as it does large banks, central banks, large corporations and
governments. This market is diffused globally through interconnected financial
actors with the three dominant centres located in New York, London and Tokyo.

The daily volume of transaction worldwide has exceeded $4,000 billion in April
2007.
Up to 20% of all business is estimated to be trading volume arising out of
commercial transactions.

The rest are interbank transactions spawned by

commercial transactions. About half the foreign exchange is for spot delivery
usually about 2 working days after dealing date. The rest are forward deals which
may be outright forwards (deals involving currencies exchange occurring one way
at a future date between 2 parties only once) or swaps (deals involving exchange of
currency between parties at a future date and a re-exchange later).
Other features which characterize the foreign exchange are its extreme liquidity,
the large number and variety its participants, its long trading hours and its low
margins of profit.

1.4 From Cash Instruments to Derivatives (Derivatives


Markets)
Theoretically the cash markets described earlier are exposed to considerable risks
because they involve the raw flows of cash. Cash transfer can be messy and may
carry with it dire consequences in the event of something going wrong and funds
not being available when they should. It would be very convenient if the risk,
which huge transfers imply, were to be managed and administered more
cautiously.

Furthermore, seeing that interest and exchange rates could be so

volatile, it would be useful if some guarantees can be bought which would serve to

10

stop rates from inflating future real prices, whose nominal value is being agreed on
at the present time. This process is called hedging.
Derivatives were designed, amongst other things, to cater for the above. Originally
commodity producers used forward and future contracts (refer to sections 1.4.1 and
1.4.2) to hedge prices and reduce risk. Today the derivatives market are linked not
only to the commodities markets, but also to underlying instruments in the cash
markets, with similar exposure to currency fluctuations, interest rates and market
swings.
Formally, a derivative contract is an agreement involving usually 2
counterparties in which they agree to transfer an asset or amount of money on
or before a specified date at a specified price. A derivatives value changes
with changes in one or more underlying market variables, such as interest
rate or foreign exchange rates.
Besides commodities, derivative trading involves all financial instruments. Basic
derivative instruments include:
o Forward contracts
o Future contracts
o Options contracts
o Swaptions
Derivative instruments are also sometimes called contingent claims as their
payout to the holder at maturity is contingent on the price or rate of the underlying
asset. Their major objectives may be summarized as:
o Lower international funding costs
o Get better exchange rates in international markets
o Hedge risk prices
11

o Diversify funding and risk management


Seeing that the idea behind derivatives starts off with a party exposed to a financial
risk which it wants to pass on to another party willing to accept, there should be at
least 2 actors involved. There are hedgers; dealers who want to pass the risk of
future price or interest rate fluctuations to others and buy guarantees for future
prices and rates. They hope that by changing the risk profile smartly they assure
more secure expected, long run returns. Speculators take on the opposite position
to hedgers, by taking on exposure to risk in the hope of making gains through
movements in prices, rates and indices. There are also Arbitrageurs who may
trade in derivatives with a view to exploit differences in prices within different
derivative markets or between derivative instruments.

1.4.1

Forward Contracts

A Forward Contract is an agreement to buy or sell a specified quantity or


quality of an asset (for ex. a commodity) at a specified future date, for a
certain price which is fixed at the present date. This price is said to be the
forward price. The investor buying the asset is said to enter a long forward
position, whereas the investor who agrees to sell the asset is said to enter a
short forward position.
Normally this contract is not traded on the exchange.

1.4.1.1

Forward Exchange and Interest Rates

With floating exchanges rate and the increasing range of borrowing facilities
offered by banks during the 60s and 70s, various financial institutions had to

12

finance themselves by borrowing short and lending long (i.e. borrowing for short
term, maybe 3 months, and lending out that same money long term, maybe 20
years. Profit generates from paying less interest on the short end than what you are
charging on the long end) in increasing quantities. The problem was that they were
forced to pay at the going rate which may over time no longer be favourable.
Interest rates became more volatile during the 70s and 80s and corporate
treasures sought bank protection from high borrowing costs. So a problem with
great theoretical implication arises: what is the fair rate to be quoted?
Many studies concentrated on the question of how well, or badly, do existing
predictors of future spot rates perform. Many theoretical models were developed
to obtain formulas which give estimates for future values of interest and exchange
rates.

It became soon evident that expected future rates computed using

forecasting techniques match rather badly with actual spot rates.

More

sophisticated models were proposed and studied to give better predictive tools. In
effect, one of the more successful methods used when banks quote forward
exchange and forward interest rates uses the principle of risk-free arbitrage. In
using this technique a dealer considers how to hedge the resulting position using
other transactions whose price is known. Having obtained an estimate of the best
rate to quote for future reference, derivative instruments are issued using this
information.

1.4.1.2

Forward Rate Agreements

A Forward Rate Agreement is an agreement between two counterparties who wish


either to hedge against or to speculate on a movement in future interest rates. It is
a forward-forward loan granted at a fixed interest rate without the actual lending

13

commitment. The seller agrees notionally to lend a particular sum to the


buyer of a specified size, in a specified currency, to be drawn at a particular
future date at a fixed rate of interest. The agreement lasts for a specified
duration.

In practice what happens involves no direct borrowing or lending.

Usually the buying company enters some future borrowing agreement and needs a
guarantee against future varying rates.
So it comes into agreement with another company that if rates exceed a fixed
number the buying company is paid the extra money it needs to pay in interest, if
the rates go down it gives the money it saves to the other company. Forward
contracts are typically quoted as a two-way price with bid-offer rates. Prices are
often quoted in standard terms, whole months and rounded numbers for notional
size.

1.4.1.3

SAFE

A Synthetic Agreement for a Forward Exchange (SAFE) is an agreement


between two parties to execute a national forward-forward exchange foreign
currency swap between a pair of currencies. The exchange is from a primary
currency to a secondary currency for specific amounts at particular spot and swap
exchange rates. The Buyer notionally agrees to buy an agreed amount of primary
currency on settlement date.

On maturity date the seller busy in secondary

currency at the agreed rate.

1.4.2

Futures Contracts

14

A Futures Contract is a firm contractual agreement between buyer and seller for a
specified asset on a fixed date in the future. The contract price will vary according
to the market price but is fixed when the trade is made. The contract has standard
specifications so both parties know exactly what is being traded. Futures are
distinguished from generic forward contracts in that they contain standardized
terms, trade on a formal exchange, are regulated by overseeing agencies and are
guaranteed by clearinghouses.
The risk to the holder has to be taken fully, and because the payoff pattern is
symmetrical, the risk to the seller is unlimited as well. Money lost and gained by
each party on a futures contract are equal and opposite. Thus futures trading is a
zero-sum game. One partys losses are the other partys gain.
Futures contracts originate from all over the world and converge on a pit which
concentrates trading in a single crowded location. Members of futures exchange
have a desk on the exchange with whom clients can transmit their orders. These
orders are transferred to floor brokers who trade contracts as their clients instruct.
The features characterizing trading at the pit are designed to emphasize the
transparency of the whole process. An alternative to the pit is trading by screen.
Services like Reuters Monitor provide prices and deals are executed by telephone.
Institutional problems of establishing a physical futures market with sufficient
liquidity have resulted in a trend in favour of smaller markets with screen-based
infrastructure.
After a trade is executed, the details are passed to the clearing house, which
ensures the correctness of the deal and removes counterparty risk by coming
15

between seller and buyer. The clearing house serves as counterparty to both sides
of the trade.
The contract is not directly between the buyer and seller: the clearing house takes
on the credit risk should a counterparty default.
The great bulk of futures contracts consist of interest rate futures, with bond and
stock index futures forming a smaller market.

Interest rate futures are forward transactions with standard contract sizes
and maturity dates which are traded on a formal exchange.

1.4.2.1

Short-term and Long-term Interest rate futures

Short-term interest rates futures are futures contract on a short term interest
rate. They are by far the most heavily traded. The three-month U.S. Treasury bill
contract, introduced by the International Money Market in 1976, was the first
futures contract based on short-term interest rates. Three-month Eurodollar time
deposit futures, now one of the most actively traded of all futures contracts started
trading in 1981. Transactions are notional fixed-term deposits where the price is
the fixed rate of interest applying during the term of the deposit covering a certain
period in the future. Cash settlement rather than asset delivery is usually the rule.
Delivery occurs on pre-defined delivery dates, usually 4 times a year. Interest
rates are traded on an indexed price. This is not a real price but a quotation system
which reverses the behaviour of future prices. If rates increase, prices decrease and

16

vice versa. Short-term interest rates are almost exclusively based on Eurocurrency
deposits and are cash settled on Exchange Delivery Price or last price traded.

Long-term interest rates futures are essentially longer-maturity interest rates


forward contracts in notional underlying fixed coupon debt instruments.
These instruments require the delivery of an actual bond and are settled on the
basis of a basket of deliverable government bonds or notes.
The settlement price is based on the final futures contract price. Settlement is by
physical delivery of bonds or notes with coupon rates and maturity dates stipulated
by this exchange.
There may be a problem if the delivery of bonds were restricted solely to those
with the specified coupon. There might be no bonds of the given coupon available.
Furthermore insistence on the same type could allow a group of investors to obtain
a major portion of the underlying bonds, and hence the ability to manipulate prices.
To overcome this problem clearing houses define a basket of deliverable bonds
which they will accept or deliver. Deliverable grade bonds will meet a set of
criteria covering their liquidity, maturity, range, coupon payments etc.

1.4.3

Options Contracts

The financial tools we have discussed so far contribute a lot towards managing risk
efficiently. However, once they are traded there is no way to go back except
possibly through trading any remaining debt instrument in ones possession. It
would be much more convenient if commitment to buy or sell would not be so
complete, but conditional on the persons wish when she so decides.

17

Options, in a sense, offer this opportunity the possibility not to buy or sell should
market conditions make these actions not desirable at the moment, or the reverse.
Options contracts were introduced on an exchange in 1973 and since then their use
exploded.

An Options contract confers the right, but not the obligation, to buy (call
option) or sell (put option) a specific underlying instrument or asset at a
specific price the strike or exercise price up until or on a specific future
date the expiry date. The price to have this right is paid by the buyer of the
option contract to the seller as a premium.
Options contracts come in various styles, however the most common contracts are
the European and American options. European style options cannot be exercised
before maturity, while American style options do not impose any conditions on
when the right to exercise is activate. The intrinsic value of an option is the net
positive amount that an option would realize if it were exercises immediately. In
virtually all cases, the option seller will demand a premium over and above the
intrinsic value. This intrinsic value may be different tomorrow and the asymmetry
of options means that the seller stands more to lose than gain. This excess is called
time value.
Interest rate options were first introduced in the 1980s to hedge interest rate
exposure. Interest rate options can be written on bonds and bond futures. Options
written on bonds give the buyer the right, but not the obligation, to buy or sell a
specified amount of bonds or notes for a pre-determined price. They are traded
both over-the-counter and on an exchange. They are usually cash settled.
18

Options on bond futures have notional government bonds as underlying


instruments. They are exchange traded and settled using the same conditions just
mentioned. They are standardized in terms of underlying instruments and amount,
strike prices, expiry prices, style, premium quotations, margin payments.

1.4.4

Swaps

Swaps are another outstanding success story from the 1980s. Swap transaction is
the simultaneous buying and selling of a similar underlying asset or obligation
of equivalent capital amount where the exchange of financial instruments
provides both parties to the transaction with more favourable conditions than
they would expect otherwise.

There are two main types of swap products:

interest-rate and currency swaps.

An interest rate swap is an agreement between counterparties in which each


party agrees to make a series of payments to the other at agreed future dates
till maturity of the agreement. Each partys interest payments are calculated
using different formulas by applying the agreement terms to the notional principal
amount of the swap.
Interest rate swaps are the most important of over the counter swap derivatives.
Also they are the most popular because they allow both parties to access better
interest rates than they would otherwise receive in the markets. The interest
amounts are calculated for both parties from the same notional principal and so no

19

physical exchange is necessary. The two rates of interest are calculated in the
same currency and the payments are netted so that only differences are settled.
The interest payment structure can vary. The most common is the plain vanilla
which secures a fixed rate for one party and a floating rate for the other. Banks
and corporations are the main market players in interest rate swaps. Different
institutions have different credit ratings and the one with lower rating will get
better terms for floating rates, while high credit ratings make fixed rates more
favourable.

A currency swap is an agreement between two counterparties in which one


party makes payments in one currency and the other in a different currency
on agreed future dates until the agreement maturity. It is a long term over the
counter derivative usually consisting of
Spot exchange of principal at the beginning
Exchange of interest rate payments at agreed dates in 2 different currencies
Re-exchange of principal on agreement maturity made at the original spot
rate.
There are many variations in the structure of exchange and payments of swap
transactions. These have evolved over the last 20years and are being designed in
response to financial needs of various institutions.

20

Chapter 2
Single Period Financial Markets
2.1 Introduction
In this chapter we initiate the task of developing a mathematical model for
capturing the workings of the securities markets and the fortunes of an investor
who is buying and selling assets over time. Such a model would display its use, if
first and foremost manages to describe, with an acceptable accuracy, real markets.
This is very hard task which, in spite of remarkable progress especially in the last
25 years, is currently very far from completion. Some very powerful, and difficult
mathematics, has been harnessed into studying the problems involved. Notably
stochastic analysis has made significant contributions.
We shall only touch on some of the problems involved and in this chapter we
simplify the setting by considering changes involving only one period.

2.2 Dynamics of Prices of Risk-free and Risky Assets

Consider a set of N+1 securities. The first security is assumed to be a riskless asset
(such as a savings account), which guarantees on giving fixed interest payments.
The other N asset could be bank deposits, treasury or trade bills, government or
corporate bonds, equities involving stocks and shares. Some yield regular interest
payments others are exposed to market risks. We need to model the set of prices as
it varies over time.

21

Let t denote time, which will be measured in discrete units (which are commonly
referred to as ticks). We shall as a first step simplify matters by taking one step,
that is t = 0 to t = 1 .
Markets are unpredictable. However their behaviour is not entirely erratic but
hopefully governed by probabilistic laws which the mathematical finance seeks to
model. Thus we suppose that open to the market for the future time there are a
number of possibilities.

We shall denote by the set of these possibilities which we can look at as possible
paths of evolution over time of the assets under consideration. As we consider
more complex and realistic models will become a larger and more structured set,
but mathematically more difficult to deal with. We will need to endow this set with
the structure of a probability space which will describe mathematically the
probabilistic laws which the market obeys.
We shall next concentrate on the values of various securities. The simultaneous
value of these securities, called a price vector, may be represented by a vector St of
asset prices which at time t, will vary according to which point in we have. In
fact St is a random vector:
T

St = ( Bt , St1 ,...StN ) , t = 0,1

We are interested in following the prices of these N+1 assets. The price, at time t,
of the riskless asst, will be denoted by Bt. This gives fixed interest payments at an
interest rate r, which is virtually assured till maturity. We shall take B0 = 1 and
Bt +1 = (1 + r ) Bt .
22

Stn ( ) will denote the price at time t of the nth risky asset when possibility

occurs, where n = 1,2,N.

Example 2.1
Suppose the financial market is composed of 4 assets (i.e. N = 4), and we move one
step forward. The first asset we assume is a bank deposit growing by 0.05 from the
value of 1. The second asset is undergoing a very calm period and will stay at 2.5.
The third asset starts at 1.8 can either grow by 0.1 with probability 0.3 or diminish
by 0.1 with probability 0.2. The fourth asset increases by 0.2 with probability 0.4
and decreases by 0.2 with probability 0.2. This gives us 9 possible future price
vectors (1 , 2 ,...9 ) as shown in the table:

Possibilities for the next step t = 1


t =0

1.05

1.05

1.05

1.05

1.05

1.05

1.05

1.05

1.05

2.5

2.5

2.5

2.5

2.5

2.5

2.5

2.5

2.5

2.5

1.8

1.9

1.9

1.9

1.8

1.8

1.8

1.7

1.7

1.7

3.6

3.4

3.8

3.6

3.4

3.8

3.6

3.4

3.8

3.6

The first two assets can only change to 1 value, the third and fourth can each have
3 possibilities : grow, stay as they are or go down. Assuming that these movements
for the two assets are independent of one another, we calculate the probabilities:

23

P {(1 )} = 0.3 0.2 = 0.06, P {(2 )} = 0.3 0.4 = 0.12,


P {(3 )} = 0.3 0.4 = 0.12, P {(4 )} = 0.5 0.2 = 0.1,
P {(5 )} = 0.5 0.4 = 0.2, P {(6 )} = 0.5 0.4 = 0.2,
P {(7 )} = 0.2 0.2 = 0.04, P {(8 )} = 0.2 0.4 = 0.08,
P {(9 )} = 0.2 0.4 = 0.08

We can work out expected values for all assets into the next period. Assets 1 and 2
we already know will value at 1.05 and 2.5. For asset 3 the expected value is 1.81
and for the 4th asset we have 3.64.
When we compare these values with the ones at t=0 we see that on average prices
for the last 2 assets will go up. Holders of such assets will not part with them
unless they are offered more than the returns they expect from them. Furthermore
they can consider the option of finding which assets give the highest expected
return and stack all their cards on this asset. But is this a good strategy seeing that
we have nothing guaranteed 100%? Also how are we supposed to accept the
probability figures quoted?
If things were so easy to quantify, why arent all persons participating in these
markets getting continuously rich by putting their money where there are gains to
be made? Some of them are very rich. Others are also stupid and so quickly weave
their own ruin. But most are neither one thing nor the other. We need to understand
better what goes on.
We first need to define mathematically the process of buying and/or selling such
securities in varying amounts over time. This is the subject of the next section.

24

2.3 Trading Strategies and Value Process


T

An investors portfolio is given by a vector H t = ( H t0 , H t1 ,...H tN ) where

H t0

refers to the number of units held in the riskless asset during the period t-1 t,
while for n = 1,2,N, H tn gives the number of units held in the nth risky asset
during period t-1 t. In general, the process H := { H t }tT , where T = {1, 2,...} is
is said to be a trading strategy, however since we are in a single period market, we
shall only consider the number of units carried forward from time 0 to time 1.
Hence, to simplify notation, we shall remove the dependency on t and write
H n to represent the number of units held in the nth risky asset during period

0 1.

H together with St gives us the information about the investors wealth Vt H at time

t:
N

Vt = H Bt + H n Stn .
H

n =1

The process V H = {Vt H }

t{0,1}

is said to be the value process.

Example 2.2
The following table gives an investors portfolio:

Hn

St in

H 0n Stn

Riskless Asset

3000

3000

Bonds

1200

1.5

1800

Stocks in Comp A

200

2.3

460

25

Stocks in Comp B

300

4.5

1350
6610

Total Value Vt H

At the beginning of each period an investor will decide how much money to keep
invested in each of the assets till the end of the coming period.
Till the end of the next step various possibilities are possible, each having its effect
on the securities prices. The price vector will now become St+1. We need
something which measures the performance of the portfolio for each possibility
and we use the value function Vt H defined earlier.

Given a trading strategy H, the value of the portfolio at time t+1, as viewed at t,
will now be a random variable dependent on future prices and is given by:
H
t +1

( ) = H

Bt +1 + H n Stn+1 ( )
n =1

2.4 Gains Process


In a single period financial market the Gains Process Gt is another related random
variable defined as follows:
N

H
t +1

( ) = H ( Bt +1 Bt ) + H n ( Stn+1 ( ) Stn )
0

t =0

n =1

This random variable gives the total profit or loss generated by the respective
portfolio.

26

2.5 Discounted processes


To compare the prices of various assets it is important to work with normalized
(discounted) prices. A typical approach is to assign to each price its todays value,
and compare these values, as this would allow us to choose today between two
different future payments. We define the discounted price vector over the start and
end of a single period by using the discount factor captured by the riskless asset:
T

S = ( B , S , S ,...S
*
t

*
t

1*
t

2*
t

N* T
t

S1
SN
= 1, t ,... t
, t = 0,1
Bt
Bt

with corresponding discounted value and discounted gains processes:


N

Vt

H*

= H + H n Stn*
0

n =1
N

GtH* = H n Stn*
n =1

where Stn* ( ) = Stn* ( ) Stn*1 . Note that Vt H* =

Vt H
= Vt H1* + GtH *
Bt

Example 2.3
T

Consider a 3-asset portfolio with corresponding price vector ( B0 , S01 , S02 ) having
four possibilities in the future t =1. B0 corresponds to a riskless asset and moves up
by 0.1 in all cases, but S01 and S02 can move only 0.1 up or 0.1 down with
probabilities as shown in the table below:

27

t=1
Asset

t=0

Bt

1.1

1.1

1.1

1.1

St1

2.3

2.4

2.4

2.2

2.2

St2

3.4

3.5

3.3

3.5

3.3

0.3

0.1

0.2

0.4

Probability

We consider the portfolio ( H 0 , H 1 , H 2 ) = (1000,500,600 ) .


T

With this portfolio we have,


V0H = 4190
V1H (1 ) = 4400 giving a gain of 210, with probability 0.3
V1H (2 ) = 4280 giving a gain of 90, with probability 0.1
V1H (3 ) = 4300 giving a gain of 110, with probability 0.2
V1H (4 ) = 4180 giving a loss of 10, with probability 0.4

So on average for the next step we expect the value of the portfolio to be 4280. The
table below gives us the discounted price vectors.

t=1
Asset

t=0

Bt*

1.0000

1.0000

1.0000

1.0000

St1*

2.3

2.1818

2.1818

2.0000

2.0000

St2*

3.4

3.1818

3.0000

3.1818

3.0000

28

So the discounted values of the portfolio under each possibility are:


V1H* (1 ) = 4000
V1H* (2 ) = 3890.9
V1H* (3 ) = 3909.1
V1H* (4 ) = 3800

We see here that there is nothing here to be gained by investing in the risky assets.
And the situation looks silly!

2.6 Arbitrage
There are conditions which the price vector St needs to satisfy if it is not to lead to
untenable situations in a competitive market. We give a simple example:
Example 2.4
Consider the evolution of the price vector over a single step as given by the table
below:
t=1
Asset

t=0

Bt

1.1

1.1

S t1

2.3

2.5

2.4

S t2

1.3

1.5

1.4

0.5

0.5

Probability

Consider, for example, the portfolio

(H

, H 1 , H 2 ) = ( 23,10,0 ) , which
T

corresponds to getting rid of 23 units from the first asset and keeping 10 units from

29

the second asset. Its value is 0 at t=0. When carried over to period t=1 this
portfolio gives a sure decrease: either 0.3 or 1.3.
T

Strategy ( H 0 , H 1 , H 2 ) = ( 13,0,11) , on the other hand, will always increase. If


T

such situations were possible, who would be willing to trade with the person
holding the second portfolio? No competitive sort of market would allow for long
one participant to make sure gains. We must now go beyond the simplicity of the
situation above and formulate the condition in a suitably general way.
Assumption 2.5
We shall assume that our probability space has a finite number of
possibilities {1 ,...K } .
Definition 2.6
A trading strategy H is said to be dominant over the period t = 0 to t =1 if there
exists trading strategy H with
V0H = V0H but V1H ( ) > V1H ( ) for all

The above condition is equivalent to the existence of a strategy creating wealth out
of nothing with certainty:
Theorem 2.7
The following statements are equivalent:
1. a dominant trading strategy exists
2. there exists a trading strategy H with V0H = 0 and V1H ( ) > 0, for all
3. there exists a trading strategy H with V0H < 0 and V1H ( ) 0, for all .

30

Proof:
(1)(2): Suppose that there exists a dominant strategy H and let H be a
dominated strategy. Then clearly H := H H satisfies the condition of the
theorem.
(2)(1): Let strategy H satisfy V0H = 0 and V1H ( ) > 0 , . Then if we take
the strategy given by the vector 0, we see that H dominates it.

(2)(3): Suppose there exists trading strategy H satisfying V0H = 0 and


V1H ( ) > 0 , . So,
N

H + H n S0n = 0

....( i )

n =1

and for all


N

H 0 B1 + H n S1n ( ) > 0 ...( ii )


n =1

Multiplying both sides of (i) by B1 we obtain


N

B1 H 0 + H n S0n = 0
n =1

...( iii )

Subtracting (iii) from (ii) gives

H ( S ( ) B S ) > 0
n

n
1

n
1 0

n =1

Letting
N

H n ( S1n ( ) B1S0n )

n =1

B1

= min

31

guarantees that > 0. We shall construct a new strategy H dominated by our H .


We leave the number of units in the risky assets the same, i.e.
H n = H n for all n = 1, 2,...N . We let the number of units in the riskless component

be equal to
N

H = H n S0n
0

n =1

Then,
N

n =1

n =1

V0H = H n S0n + H n S0n = < 0

Also,
V1H ( )
N
N n n

= H S0 B1 + H n S1n ( )
n =1
n=1

= H n ( S1n ( ) B1S0n ) B1
n =1
N

= H
n =1

( S ( ) B S ) min H ( S ( ) B S ) 0
n
1

n
1 0

n
1

n
1 0

n =1

(3)(2): If the strategy H satisfies (3), reversing the argument above, setting
N

H = H n S0n and leaving H n for all n = 1, 2,...N , the same, we get strategy H
0

n =1

satisfying (2).
We shall give yet another equivalent definition to the one above. This definition
will prove to be extremely useful later on, when dealing with pricing of options
contracts. It says that given future prices, there exists a probability measure

32

with respect to which the average discounted future value of a portfolio is


equal to the current value of that same portfolio.
Definition 2.8
The pricing of future claims is said to be consistent if there is a vector of nonT

negative components = ( (1 ) ,... ( K ) ) , which we call linear pricing


measure, such that for every trading strategy H we have:
K

H
0

= V1
i =1

V1H (i )
(i )
(i ) (i ) =
B
i =1
1
K

H*

We relate the above to a probability measure:


Theorem 2.9
T

Given a vector = ( 1 ,... K ) , the following conditions are equivalent:


1. is a linear pricing measure with (i ) = i
2. is a probability measure on satisfying:
K S11 (i ) (i )

B1
i =1
1*
1

S0 S 0
2
2* 2 K S1 (i ) (i )

S0 S 0
=
=
B

i =1
1


S N* S N

0 0 K SN
1 ( i) ( i)

B1
i=1

Proof:
(1)(2)
If theres a linear pricing measure then, by definition, for any strategy H
33

H
0

= V1

V1H (i )
(i )
(i ) (i ) =
B1
i =1
K

H*

i =1

Let us take in particular the strategy with H 0 =1 and H n =0 for all n = 1,2,...N so
thatV0H = 1. But
K

H
0

= V1
i =1

V1H (i )
(i )
(i ) (i ) =
B1
i =1
K

H*

so
K

K
B1
V = (i ) = (i )
i =1 B1
i =1
for this portfolio. Since, by definition, has non-negative components, it can be
H
0

considered as a probability measure. To get the equations in (2), for arbitrarily


chosen j {1, 2,...N } , let
N T

( H ,...H ) = ( 0,..,0,1,0,...0 )
1

where the unit 1 occurs in the jth position. Assume also that H 0 = 0 . Then
N

V0H = H 0 + H n S0n = S0j


n =1
K

(i )

i =1

B1

(i )

i =1

Bi

n n
H B1 + H S1 (i )
n =1

S1j (i )

(2)(1)
Suppose that the is form a probability measure which for 1 n N satisfies the
condition :
K

(i )

i =1

Bi

S =
n
0

S1n (i )

It follows then that for any strategy H:


34

H 0 + H n S0n
n =1
N

= H + H
0

(i ) S1n (i )

i =1

B1

n =1

0 N n S1n (i )
= (i ) H + H

B1
i =1
n =1

But
N

H + H n S0n = V0H
0

n =1

And
S1n (i ) V1H (i )
H + H
=
B1
B1
n =1
N

Hence we have shown that is a linear pricing measure.

We finally establish contact between linear pricing measures and dominating


strategies:
Theorem 2.10
The following two conditions are equivalent:
1. is a linear pricing measure
2. No dominant trading strategy exists
Proof:
Recall from Linear Programming, the following table which shows how to convert
a primal to a dual modal

35

Variables

Constraints

Minimization
Problem
0
0
Unrestricted

Maximization
Problem

0
0
Unrestricted

Constraints

Variables

From the above table we know that the following problems are 2 dual LP
problems:

PROBLEM 1 :

max ( 0,0,...0 ) 2



K
st
1
1

S 1*
1*
1 ( 1 ) S1 (2 )

S N * ( ) S N * ( )
1
1
2
1
0

PROBLEM 2 :

1
1
S (K ) S 1
2 0
=


N
N*
S1 (K ) K S0

1*
1

H10
1
1
N H1
min (1, S0 ,...S0 )

HN
1
st
1 S11* ( 1 )

1*
1 S1 (2 )

1*
1 S1 (K )

0
S1N * (1 ) H 0

S1N * (2 ) H 1 0

S1N * ( K ) H N 0

36

Problem 1 captures the condition that decreasing future be consistent with all linear
pricing measures.
Problem 2 tries to find the smallest portfolio which gives positive values under
all possibilities.
(1)(2)
Suppose condition (1) holds and consider Problem 1. By Theorem 2.9, is a
probability measure which satisfies (2) in theorem 2.9. So problem 1 admits a
solution and its optimal objective value is clearly zero.
Recall the Strong Duality Property, again from Linear Programming:
If one problem possesses an optimal solution, then both problems possess optimal
solutions and the two optimal objective function values are equal.

By the above result the optimal value objective value in problem 2 is zero. But the
objective function in problem 2 is precisely V0H (so V0H = 0 ). On the other hand the
constraints represent V1*H (so V1*H ( ) 0 for all ).
This means that there cannot exist a strategy with V0H < 0 (since the minimal
V0H is zero) and V1*H ( ) 0 for all , which, by Theorem 2.7, is equivalent to

the absence of dominant trading strategies.


(2) (1)
Suppose there were no dominant strategies. Then with the constraints
1 S11* (1 )

1*
1 S1 (2 )

1*
1 S1 (K )

0
S1N * (1 ) H 0

S1N * (2 ) H 1 0

S1N * (K ) H N 0

37

(or equivalently V1H ( ) 0 for all ), V0H cannot be less than zero.

So

Problem 2 admits a solution and, clearly, the vector H = 0 is an optimal solution to


this problem.
Again by duality theory it follows that Problem 1 also admits an optimal solution
which can be taken as the linear pricing measure (as mentioned in the first part of
the proof). Hence the existence of this measure is established.

The conditions we have just discussed, involving the existence of a dominating


strategy, is in fact rather strong. It gives an extremity which should never ever be
allowed. But there is a weaker condition which should also be ruled out. This is
given by the following definition:
Definition 2.11
A trading strategy H is said to be an arbitrage opportunity if
i. V0H = 0
ii. V1H ( ) 0
iii.

E V1H > 0

Such a strategy would start with value 0, go through a period without


exposure to any risk at all of losing money, and on average make strictly
positive gains. Possibilities of this nature should be excluded. We first see that
this condition is weaker than the existence of a dominant strategy.
Theorem 2.12
Existence of a dominant trading strategy implies the existence of an arbitrage
opportunity. However there could exist an arbitrage opportunity but not dominant
strategy.
Proof:

38

Suppose there exists a dominant strategy H. Then by theorem 2.7 there exists a
trading strategy H with V0H = 0 and V1H ( ) > 0 for all . Clearly for this
strategy E V1H > 0 , as required.
We next give a counterexample to prove the second part of the theorem. We shall
give an example where an arbitrage opportunity exists and yet no dominating
strategy can exist.
Consider the following 2-asset prospectus:
t=1
Asset

t=0

Bt
S t1

2.3

2.4

2.3

1-p

Probability

We see that the strategy H = (-23,10)T is an arbitrage opportunity. Yet = (0,1)T is


a linear pricing measure whose existence rules out the presence of a dominant
trading strategy.

We shall now obtain a condition, which is equivalent to the possibility of arbitrage,


but easier to deal with. This condition is a slight strengthening of the linear pricing
measure condition.

The latter excluded the existence of a dominant trading

strategy but not arbitrage opportunities.

2.7 Risk Neutral Measures


Given probability measures P1 and P2 on the same measurable space ( , F ) we
can work out different averages for the same random variable X defined on the pair

39

above. They will not necessarily be equal. We denote the expectation of X with
respect to P1 by E P1 [ X ] and similarly with respect to P2 . We present this sections
important concept:
Definition 2.12
Given a set of future possibilities on which we define a probability measure Q .
Then Q is said to be a risk neutral probability if:
Q {} > 0 for all
E Q S n* := E Q S1n* S0n = 0 for each n = 1,2,...N

Example 2.13
Consider a financial market composed of two risky assets and a riskless asset. In
general there will be much more future possibilities than assets and so satisfying
the corresponding set of risk neutral equations may be difficult for we have too
many equations to satisfy. For simplicity we assume that there are only three future
possibilities 1,
2 T
0

2 ,3. Let the current price vector be given by

( S , S ) = (10, 25)
1
0

. Let us consider 3 different sorts of future discounted prices,

given in matrix form as shown below:


Case 1:
1 1 1 p1 1


12 10 8 p 2 = 10
15 20 30 p 25

40

The first row corresponds to the property of a probability measure (i.e. sum to
T

one). There is a unique solution to this : ( p1 , p2 , p3 ) = (1, 1,1) which clearly does
not give us a probability distribution.

Case 2 :

1 1 1 p1 1


12 10 8 p2 = 10
24 20 16 p 25

3
This set of equations is really just 2 equations which can be reduced to
p3 = 1 p1 p2 and p2 = 1 2 p1 so that there are 2 solutions which exclude risk
T

neutral

probability

( p1 , p2 , p3 )

1 1
= ,0,
2 2

and

( p1, p2 , p3 )

= ( 0,1,0 )

remembering that the risk neutral measure must be strictly positivewhile the
general solution is (, 1-2, ) for 0 < < 1. Situations like this can be found
which exclude completely probabilistic interpretations for the ps.
Case 3:

1 1 1 p1 1


12
10
8

p2 = 10
23 20 30 p 25

3
The above matrix shows the existence of a risk neutral measure which is given by
the unique solution:
T

( p1, p2 , p3 )

= ( 0.38462,0.230769,0.384615 ) .

41

Existence of a probability measure as described above is precisely the workable


condition we need to exclude arbitrage opportunity. We see this in the following
theorem:

Theorem 2.14
The following 2 conditions are equivalent:
1. There exist no arbitrage opportunities
2. There exists a risk neutral measure
Proof:
(1)(2)
Form a set W of all random vectors corresponding to the gains vector in future of
some portfolio H:
W = { X = ( X 1 , X 2 ,... X K ) : G1H* (1 ) = X 1 ,...G1H* (K ) = X K for some H }

Then W is a vector subspace of RK (it is not difficult to show that for all X , Y
W and for all real numbers , we have X + Y W Exercise).
Let O+ be the positive orthant of RK without the origin, that is
O + = {( x1 , x2 ,...xK ) : xi 0, x1 + x2 + ...xK > 0}

Saying that there does not exist any arbitrage opportunities means there exists no H
such that
V0H = 0
V1H ( ) 0 , that is G1H* ( ) 0,

42

B1G1H* > 0 (It is not difficult to see that


E V1H > 0 , that is E
V1H = B1G1H * - Exercise)

But this is equivalent to saying that W O+ = .


Now consider the set
A+ =

{X = ( X , X
1

,... X K ) : G1H* (1 ) = X 1 ,...G1H* (K ) = X K for some H with E [ X ] 1} O +

It is not difficult to prove that A+ is a closed and convex subset of RK Exercise.


Also, in the absence of arbitrage opportunities, this set is disjoint from W
Exercise.
Recall, from Non-Linear Programming, the Separating Hyperplane Theorem:
Let S1 and S2 be two non-empty, closed, convex and disjoint subsets of K . Then
there exists p n , p 0 such that
inf { pT x : x S1} sup { pT x : x S 2 }

Let S1 = W and S2 = A+ . The above result is equivalent to saying that there exists
a K-dimensional vector Y which lies in the orthogonal complement of S1 such that
for all XS2 X.Y > 0.
However it is clear that for any selected index j, we can always choose vector X
from A+ which has 0 components except for the jth which must be strictly
positive. This forces all Y components to be strictly positive. The required risk
neutral probability is then given by:

43

Q { j } =

Yj
Y1 + ...YK

(2)(1)
Suppose risk neutral probability measure Q , that is
Q { j } > 0 for all
E Q S n* := E Q S1n* S0n = 0 for each n = 1,2,...N
Take any trading strategy H, we have
N
N
E Q G1H* := E Q H n S1n* = H n EQ S1n* = 0
n=1
n=1

So there cannot exist G1H* , such that G1H* ( ) 0 for all and E Q G1H* > 0 .
But this is equivalent to not having arbitrage opportunities Exercise.

Example 2.15
Consider a 2 asset situation with price vector profile:
t=1
Asset

t=0

St1
St2

10

12

giving the gains profile:

44

S n*

St1

-1

St2

-2

-2

We have in this case:

W = X = ( X 1 , X 2 ,... X K ) : G1H* (1 ) = X 1 ,...G1H* (K ) = X K for some H = ( H 1 , H 2 )

{( H

1
1

+ 2H 2 , H 1 2 H 2 , H 1 2 H 2 ) , H 1, H 2

This vector space is really two dimensional and can be described as:
W = {( X 1 , X 2 , X 3 ) : X 1 + X 3 = 0}

The orthogonal complement of W is one-dimensional and given by:


W = {( x,0, x ) : x }

The intersection of W with the positive orthant is given by:


W O + = {( 0, X 2 ,0 ) : X 2 > 0}

This shows that there is an arbitrage opportunity. Let us see which portfolio will
give us gain zero for 1, gains X2 for 2, and gain 0 for 3. It will be the portfolio
satisfying:
H 1 + 2 H 2 = 0 and H 1 2 H 2 = X 2

This solves to H 1 =

1
1
X 2 and H 2 = X 2 . If we take X 2 to be any positive
2
4
T

number, say X 2 = 6 , we see the arbitrage opportunity with H = 3,


2

Indeed,

45

at t = 0 the value of the portfolio is zero, for 1 we get zero gain, for 2 we get
strictly positive gains and for 3 we get 0 gain.

2.8 CONTINGENT CLAIMS


Contingent claims usually refer to assets whose value is known to vary
probabilistically and agreements on future carrying forward, or not, executing a
deal are made contingent on some event or other. These events are usually related
to the value of some securities, happening or otherwise. We consider a simple
example to rehearse the risk neutral argument in this context.
Stock S has value 10 at t = 0. In the next instant it can go to 20 with probability
p or to 7.5 with probability 1-p.
Consider a European call option with strike price 15. This means that a person
buying this option now, will have the possibility to buy the stock at price 15 should
the person so decide. Thus the person can gain 5 units if the price goes to 20 or else
refrain from buying if the value goes down to 7.5. So the contingent claim
5 if =1
is: X ( ) =
. How much should this option cost?
0 if =2

Let us assume that the discounting factor is 0.8 (i.e. 0.8 =

1
). Then the risk neutral
B1

probabilities q, 1-q should give


10 = 0.820q + 0.87.5(1-q)
which fixes q at 0.4.

46

Then the average discounted future value of any amount of this stock with respect
to this probability measure is always equal to its original value at t = 0. The price
X
should be the average profit, that is E Q = 5 0.8 q = 1.6
B1

in t = 0 prices (in order for the two parties to be happy).


We shall see how this price is uniquely the one which makes sense. Consider the
following hedge portfolio

(H

, H 1 ) . We want a portfolio which replicates the

option (i.e. the value at time one of the portfolio coincides with the contingent
claim) and we work in t = 1 prices:
H0
H0
1
If stock goes to 20: 5 =
+ 20 H . If the stock goes to 7.5: 0 =
+ 7.5 H 1
0.8
0.8

This gives H 0 = 2.4 and H 1 = 0.4 .


Then the value of this portfolio at the beginning is given by 2.4 + 0.4 10 = 1.6 ,
which is precisely the price of the option. So the hedging strategy goes as follows:
Sell enough of the option to obtain 1.6 capital and borrow 2.4 in order to
invest 4 in the stock. This buys 0.4 shares. For the next time instant do as
follows:
If S11 ( ) = 20 the option would be exercised at a cost of 5 in t = 1 prices, the
loan is repaid at 3 (i.e. 2.4/0.8) and the shares are sold at 20 each. Net gain
in t = 1 prices is -5 - 3 + 20 0.4 = 0 .
If S11 ( ) = 7.5 the option will not be exercised, the repayment of the loan is
still at 3 and selling the shares obtains 7.5 for each share. Net gain is
0 - 3 + 7.5 * 0.4 = 0 .

The net result of all this is that with this replicating portfolio nothing is risked.
We see that selling the option and maintaining the portfolio above exactly balances
because the initial price of the option is set at 1.6. Note that the probabilities with
47

which the stock changes its values were not used at all. If we priced out options
higher we would have made a riskless profit. If it was less than 1.6 we would have
made a riskless profit by exchanging our roles and buying options rather than
selling.
When a trading strategy, called replicating portfolio, exists with this initial value
V0H , we say that the contingent claim is attainable or marketable. It could be

argued that this initial V0H , the value of our riskless portfolio should be the price of
our option. It promises no gains and no profits. However, one can point out
immediately that to argue for its unicity, we must show that no other portfolio
replicates the contingent claim with initial value different fromV0H . More
precisely:
The Law Of One Price holds for a securities market if there exist no trading
1

strategies H 1 and H 2 with V1H ( ) = V1H ( ) for all and V0H > V0H

This is a weaker condition than that prohibiting dominating strategies. This result
is easy to prove. We next show how the law of one price affects valuation:
If the law of one price holds, then the value V0 of an attainable contingent claim at
N

t=0 with replicating trading strategy H, is given by : V0 = H B0 + H n S0n .


0

n =1

48

Theorem 2.16
If there are no arbitrage opportunities then the t=0 value of an attainable contingent
claim, which can be considered as a random variable X in the future, is given by
X
EQ where Q is any risk neutral probability.
B1

Proof:
With no arbitrage opportunities we have the existence of a risk neutral probability
satisfying;
E Q S n* := E Q S1n* S0n = 0 for each n = 1,2,...N .

For an attainable contingent claim X, its value at t=0 is given by the value of a
replicating trading strategy H:
N

H
0

= H B0 + H n S0n
0

n =1
N

= H 0 + H n EQ S1n*
n =1

S1n
X
= H + H EQ = E Q
n =1
B1
B1
N

Example 2.17
Consider the following price vectors:
t=1
Assets

t=0

Bt

S t1

9/8

9/8

10

12

49

The existence of a risk neutral probability is provided by the solution of the system
of linear equations:
8 12 q1 8 9 q2
+
= 10
9
9
q1 + q2 = 1

which gives q1= and q2 = . This guarantees the existence of no arbitrage


opportunity.
We now want to check whether the following contingent claim X is attainable:
X (1 ) = 3
X ( 2 ) = 0

Then
X 3 3 8
+0=2
EQ =
B
4

9
1

Let us now see which portfolio generates X:


For 1

12 8 H 1 3 8
: H +
=
9
9
0

98 H1 08
For 2 : H +
=
9
9
0

This gives H 0 = 8, H 1 = 1 . Thus X is attainable.


X
Clearly the initial value of this portfolio is 2, which is equivalent to EQ . This
B1

should be the right price for X.


To activate it start with 2, borrow 8 at riskless rate of 1/8. Use the sum 2 + 8 =10 to
buy 10/10 units of stock. At time t=1 you pay 89/8 to settle the loan. If 1

50

happens, the value of the portfolio = 12 9 = 3. If 2 happens the value of the


portfolio = 9 9 = 0.

Example 2.18
Consider the following price vectors:
t=1
Assets

t=0

Bt

S t1

12/11

12/11

22

30

20

We want to find the price which should be charged for someone who wants to own
an option to buy the risky security at t=1 for the price of 25 (call option).
The existence of a risk neutral probability is provided by the solution of the system
of linear equations:
30 11 q1 20 11 q2
+
= 22
12
12
q1 + q2 = 1

which gives q1= 2/5 and q2 = 3/5 . This guarantees the existence of no arbitrage
opportunity.
Our contingent claim X can be written as follows:
X (1 ) = 5
X ( 2 ) = 0

because if the price goes down to 20, it does not make sense to exercise the option.
If the price goes up to 30, clearly buying at 25 give a net profit of 5. We next prove

51

that X is attainable by displaying a portfolio which replicates X,

that is the

existence of ( H10 , H11 ) such that:


For 1

12 H 0
:
+ 30 H 1 = 5
11

12 H 0
+ 20 H 1 = 0
For 2 :
11

This gives H 0 =

55
1
and H 1 = .
6
2

Then the price of the option is the value of this portfolio at t=0, that is

55
1 11
+22 =
6
2 6

From the other point of view we see that the expected value of X according to our
X
2 11
3 11
risk neutral probability is exactly: E Q = 5 + 0 = .
5 12
5 6
B1

Let us now consider a put option at strike price again 25. We let our contingent
claim, giving the right of the option holder to sell the security at price 25 at t=1, be
denoted by Y. So
X (1 ) = 0
X ( 2 ) = 5

because if the price goes up to 30, it make no sense to exercise the option, but if
the price goes down to 20, clearly selling at 25 gives a net profit of 5. Now for the
replicating portfolio:
For 1

12 H 0
:
+ 30 H 1 = 0
11

52

12 H 0
For 2 :
+ 20 H 1 = 5
11

This gives H 0 =

1
55
and H 1 = . Then the price of the option is the value of this
2
4

portfolio at t=0, that is


55
1 11
22 =
4
2 4

From the other point of view we see that the expected value of Y according to our
Y
2
3 11 11
risk neutral probability is exactly: E Q = 0 + 5 = .
5
5 12 4
B1

2.9 COMPLETE AND INCOMPLETE MARKETS


Given a contingent claim, the existence of a risk neutral probability does not
suffice for us to use the corresponding synthetic probabilities and work out a
valuation. We have to check that the contingent claim is marketable. That is we
have to check that it is possible to have it replicated with some portfolio. Once the
condition is satisfied, then we can proceed to obtain an appropriate valuation as we
saw in the examples above.

Definition 2.19
A securities market model is said to be complete if every contingent claim X can
be replicated by some trading strategy.

53

Theorem 2.20
Suppose that there exist no arbitrage opportunities within a securities market
model. Then the model is complete if and only if the number of point in the
probability space is equal to the number of independent vectors in the future
price matrix:
B1 (1 ) S11 (1 )

1
B1 (2 ) S1 (2 )

1
B1 (K ) S1 (K )

S1N (1 )

S1N (2 )

S1N (K )

Proof:
The result is easy to see since a contingent claim is nothing else but a random
variable X which gives us the random vector:
X (1 )

X ( 2 )

for which a replicating trading strategy exists if and only if a solution

X ( )
K

exists to the equation:


B1 (1 ) S11 (1 )

1
B1 (2 ) S1 (2 )

1
B1 ( K ) S1 (K )

0
S1N (1 ) H X (1 )

S1N (2 ) H 1 X (2 )
=

S1N (K ) H N X (K )

54

Example 2.21
We would like to find out whether given a securities market with 4 future states
11
10
11
10
11
10
11
10

and future price vector given by :

13

24 6 15

30 7 18

35 11 20

22

will permit the contingent claim:


X (1 ) = 0
X (2 ) = 10
X (3 ) = 5
X (4 ) = 20

The price matrix is invertible. Its inverse is given by:

2.386364

1.931818

-4.09091

0.681818

0.416667

-0.58333

0.166667

-0.125

0.375

-0.5

0.25

-0.79167

0.708333

0.5

-0.41667

Seeing that the rank is 4 and it is equal to the number of states, we can immediately
conclude that the claim is attainable. In fact to get the replicating portfolio we need
only multiply the above matrix by the claim vector to obtain:
53.04909
2.5

11.25

3.75

55

To obtain the valuation then we need to work out the value of this portfolio at t=0.
Or else, as previously, we get the risk neutral probability and average with respect
to it future gains for this portfolio.
It would be useful to express attainability in terms of valuation with reference to
risk neutral probabilities. It turns out that attainability ensures that valuation under
all such probabilities is the same, a condition we considered earlier:

Theorem 2.22
X
A contingent claim X is attainable if and only if EQ is constant for every risk
B1

neutral probability Q .
Proof:
To prove this result we first need to assume that the set of risk neutral probabilities
is non-empty.
()
X
Suppose contingent claim X is attainable. By the proof of theorem 2.16 EQ is
B1

constant with respect to any risk neutral probability measure as this is equal to the
V0H , the initial value of the replicating portfolio.

(
)
We prove this by contrapositive. So we show that if X is not attainable then we
X
X
can find Q1 , Q2 such that E Q1 E Q2 .
B1
B1

Unattainability implies that the equations below do not have a solution:

56

B1 (1 ) S11 (1 )

1
B1 (2 ) S1 (2 )

1
B1 ( K ) S1 (K )

0
S1N (1 ) H X (1 )

S1N (2 ) H 1 X (2 )
=

S1N (K ) H N X (K )

X (1 )
S1N (1 )

S1N (2 )
X ( 2 )
, X =
and consider an

X ( )
S1N (K )
K

B1 (1 ) S11 (1 )

B1 (2 ) S11 (2 )
H

Let S =

1
B1 (K ) S1 (K )

equivalent formulation of Farkas Lemma:


Let A be an N K matrix, and c a K-dimensional vector. Then exactly ONE of the
following two systems has a solution:

System 1:

Ax = 0 and cTx > 0

System 2:

AT= c

for some x in

Set A = ( S H ) so that A is an N K matrix , c = X so that c is a K 1vector and


= H so that is a N 1 vector.
Then system 2 reads
AT = c

that is
B
1 1
B
1 2

B1 K

( )
( )

( )

( )
S ( )
S11 1
1
1

1
1

( )

S K

S1N

( )
( )

S1N 1

N
1

( )
K

H 0 X ( 1 )
1
H = X ( 2 )

N
H X ( K )

Since X is not attainable this system has no solution, thus system 1 has a solution.

57

System 1 reads as follows:


T
Ax = 0 and c x > 0

that is
B
( ) B1 ( 2 )
1 1
S 1 ( ) S 1 ( )
1
2
1 1

N
N
S1 ( 1 ) S1 ( 2 )

B1 ( K ) x 0
1
1
S1 ( K ) x2 0
=


S1N ( K ) x K 0

and

( X ( )
1

X ( 2 ) X ( K )

x1
x
2 >0

x
K

Now take any risk neutral probability measure Q1 , then for each k, and > 0 the
expression Q1 (k ) + xk B1 (k ) can be made strictly positive provided is small
enough, call it k. Then take to be the smallest of all ks , so that for this ,
Q1 (k ) + xk B1 (k ) > 0 for all k =1, ... , K.

We let Q2 be defined as follows:


Q2 (k ) = Q1 (k ) + xk B1 (k )

We shall use Farkas Lemma to show that Q2 is another risk-neutral probability


measure.
Indeed, by the 1st row-column multiplication above we see that when we add all
the xk B1 (k ) s we get 0 and so Q2 is indeed a probability measure because Q1
is. Also,

58

S1n K S1n ( k )
E Q2 =
Q 2 ( k )
B1 k =1 B1
K
K
S1n ( k )
=
Q1 (k ) + S1n ( k ) xk
B1
k =1
k =1
S1n
= E Q1
B1

since the last term in the summation is 0 by the matrix equation we obtained
earlier. So Q2 is a risk neutral measure.
Given the claim X, we see that:
X K X ( k )
E Q2 =
Q 2 ( k )
B1 k =1 B1
K
K
X ( k )
=
Q1 (k ) + X ( k ) xk
B1
k =1
k =1
X
= E Q1 + xX
B1

Since, from Farkas Lemma, xX > 0, it follows that the expectations are not the
same.

Theorem 2.23
A securities market model is complete if it has one and only one risk neutral
probability.

Proof:
(
)

59

Suppose there exists only one risk neutral probability Q for a model. For any
X
contingent claim X , E Q with one value makes X attainable and the model
B1

complete
(
)
We prove this by contradiction.
Suppose that we have two distinct risk neutral probabilities Q1 and Q2 . Then on
some possibility , Q1 ( ' ) Q2 ( ') . Consider the following contingent claim X:
X ( ) = 0

if '

X ( ) = B1 ( ) if = '

Then we have
X
X
X
E Q1 = Q1 ( ' ) and E Q2 = Q2 ( ') E Q1
B1
B1
B1
X
This shows E Q is not constant as Q varies over risk neutral probabilities,
B1

contradicting completeness.

In a sense we have closed a circle of ideas which centre about the existence of the risk
neutral probability. This artificial probability allows us to value propositions one
would like to put on the market under the proviso that such propositions be replicable.
This condition is equivalent to uniqueness of the risk neutral probability.

60

2.10 RETURN
Before concluding the current chapter we introduce another useful terminology in
finance; the return process. We shall define this term relative to the prices of the
assets and the value and gains processes.
We shall start by giving a necessary definition.

Definition 2.24
Given a securities market model built on a probability space ( , F , P ) , and risk
neutral probability measure Q , the state price vector is the random variable L
defined by
Q {}
L ( )
=
P {}

Definition 2.25 (Return from a single asset)


Given the nth risky asset with future price vector S1n the return from this asset is a
random variable Rn defined by
S1n ( ) S0n
R ( ) =
.
S0n
n

Similarly, the return from the riskless asset is defined by R 0 =

B1 B0
=r.
B0

We shall do few computations with the returns to get:

61

S1n ( ) S0n ( )
S ( ) S =

B1
B0
n*
1

n*
0

S1n ( ) B1S0n ( )
=

B1
B1

(1 + R ( ) ) S (1 + R ) S
=
n

n
0

n
0

1 + R0
n
0
n R ( ) R
= S0
1 + R0

Recall that for probability measure Q to be risk neutral it has to satisfy the
equation:
E Q S1n* ( ) S0n* = 0

So in terms of the return process we have


R n R0
Rn R0
E Q S0n
=
E
= 0 for n = 1,2,...N .
Q
0
0
1
+
R
1
+
R

And since we are assuming that the interest rate is fixed at r, we get the further
simplification that
Rn r
EQ
=0
1+ r
E Q R n = r

(1)

We now derive formulas for the return random variable R.

62

Given a security n and a risk neutral probability Q, the corresponding state price
vector L plays a useful reference role as we shall see.
Directly from the definitions we have:
Q K Q (i )
E[ L] = E =
P (i ) = 1
P i =1 P (i )

and
n
n Q K R (i ) Q (i )
E R L = E R
=
P (i ) = E Q R n

P (i )
P i =1
n

Furthermore:
Cov R n , L = E R n L E R n E [ L ]
= E Q R n E R n = r E R n

where the last equality follows from (1) above.


The difference
E R n E Q R n = E R n r = Cov [ Rn , L ]

is called the risk premium for the security n. r is the risk neutral rate and thus
what we get in excess of it can be viewed as compensation for the risk assumed
when possessing the asset. Obviously this has to be positive if it is to attract
investors to undertake risk.
H

We next consider the return RV of a portfolio.


Definition 2.25 (Return from the value process)
Consider portfolio H = ( H 0 , H 1 ,...H N ) with initial value V0H > 0 . Then the return
from the value of a portfolio is given by:

63

VH

V1H ( ) V0H
( ) =
V0H

Using the previously defined relationships S1n ( ) = S0n ( R n ( ) + 1) , we get

VH

V1H ( ) V0H
( ) =
V0H
N

H 0 B1 + H n ( ) S1n ( ) H 0 H n ( ) S0n
n =1

n =1
H
0

V
N

H 0 r + H n ( ) S0n ( R n ( ) + 1) H n ( ) S0n
n =1

n =1
H
0

N
H0
H n S0n n
= H r + H R ( )
n =1 V0
V0

which we can interpret as the decomposition of each individual possible return into
fractions of investment in riskless asset and n securities multiplied by the
respective returns (in particular the return on the portfolio is a convex combination
of the return on the individual securities).
We can use this equation and repeat the analysis which we did further up
individually on each asset to deduce that :
E RV E Q RV = E RV r = Cov RV , L

...(*)

Along this line of thinking it makes sense to look at the behaviour of contingent
claims compared to an arbitrary portfolio with reference to state price vector L:
64

Theorem 2.26
Given state price vector L relative to a market prices probability space and take any
real numbers a, b such that the contingent claim a + bL is attainable by a
H

replicating portfolio with return RV . Given the return RV

of any other portfolio

we have:
Cov RV , R V

E RV H r
r =
H

V ar RV

E RV

Proof:
Fix a and b with b 0 such that a + bL corresponds to an attainable claim and
hence has a replicating portfolio H for which we have starting value V0H and final
value V1H = a + bL .

But since V0H 1 + RV

= V1H then L =

V1 a
=
b

V0H 1 + RV

)a

. Given any

other portfolio K with rate of return R := RV , we have

) = V

V H 1 + RV H a
0

Cov [ R, L ] = Cov R,

V H 1 + RV H
0
= Cov R,

H
0

Cov R, RV

b
H

H
V0H
Then (*) allows us to write : E [ R ] r =
Cov R, RV .

In particular for RV we have:


65

V
V
E RV r = 0 Cov RV , RV = 0 Var R V

b
b
H

V0H
allowing us to substitute for
giving :
b
Cov R, R V

E RV H r
E[ R] r =
H

Var R V

Cov RV , RV

is called the beta of trading strategy H with


The coefficient
H
V
Var R

return RV relative to K with return R.


Thus the risk premium of a portfolio is proportional to its beta relative to a linear
transformation of the state price density.

66

Chapter 3
The Optimal Portfolio and Optimal
Consumption-Investment Problem in Single
Period Financial Markets
3.1 OPTIMAL PORTFOLIOS
The problem at the heart of our studies in this branch of financial mathematics
surely has to be that of finding the portfolio which under the constraints of
operating securities market maximizes some measure of performance of the
corresponding portfolio. Such measures are usually modeled by utility functions.
In simple terms, a utility function reflects the preferences of an individual for
wealth and the risk he or she is willing to take to achieve a higher wealth.
Definition 3.1
A map u : is said to be a utility function if for each fixed ,
u(x,) is a concave and strictly increasing function in x. Then given a portfolio H
its expected terminal wealth utility is given by
E u (V1H , ) = u (V1H ( ) , )P ( )

If we let H denote the set of all trading strategies, we can express the problem as
finding H H whose expected utility at t =1 is equal to:

67

max E u (V1H , ) = max E u B1 (V0H + H 01S 1* + H 02 S 2* + ...H 0N S N * ) ,

H H
H H

Existence or otherwise of a portfolio whose expected utility is given as above is


called the optimal portfolio problem.
Unless specified we shall assume that the utility function is the same for all ,
so we can write it as a function of one argument
u:

Theorem 3.2
If a securities market model admits of a solution to the optimal portfolio problem
then the same model allows no arbitrage opportunities.
Proof
We shall prove this theorem by proving the contrapositive.
So suppose there is an arbitrage opportunity. Then we have a portfolio H such that
V0H = 0
V1H ( ) 0 , that is G1H* ( ) 0,
E V1H > 0
For any portfolio H' = ( H '1, H '2 ,...H 'N ) with value V0H' = 0 , we have
N

n =1

n =1

n =1

V0H' + H 'n S n* V0H' + H 'n S n* + H n S n*

Now this allows us to define a new portfolio H + H' whose initial value has not
changed, but whose expectation under the utility function u has to increase (since u
is an increasing function). But then if any H' were to be optimal, this new portfolio
would contradict it.

68

Definition 3.3
A securities market model is said to be viable if there exists utility function u for
which there exists a portfolio G such that G arg max E u (V1H ) .
H H

And now we establish the link between optimality and arbitrage opportunities
which we already linked very intimately with risk neutral probabilities:
Theorem 3.4
A securities market model is viable if and only if it admits of a risk neutral
probability measure.
Proof:
(
)
Suppose the model is viable. The existence of a solution to the optimal portfolio
problem guarantees the exclusion of arbitrage opportunities which in turn gives us
the existence of a risk neutral probability. We can actually go one step further and
compute such a probability.
The function we are optimizing may be written as:

u ( B (V

H
0

+ H 1S 1* ( ) + H 2 S 2* ( ) + ...H N S N * ( ) ) P {}

which should satisfy :

H n

u ( B (V
1

H
0

+ H 1S 1*() + .. + H N S N *()))P[{}] = 0

u ( B1 (V0H + H 1S 1*() + .. + H N S N *()))P[{}]B1S n*() = 0

B1E[ u (V1H )S n*] = 0


x

E[ u (V1H )S n*] = 0 for 1 n N


x

69

Let H with corresponding value V1H be the optimal portfolio whose existence is
our hypothesis. Then we define the probability measure Q as follows:
u (V1H ())

P {}
x
H

(u (V1 ))
E

Q {} =

Positivity of Q follows immediately from the fact that u is an increasing function.


By summing over all the we see that we are averaging with respect to the
probability P and it is indeed equal to 1.
Finally we need to show that it satisfies:
E Q S n* = E Q S n* S0n = 0 for n = 1, 2,...N
But:
E Q S n*

u (V1H ( ) )
= S

n*

( )

P {}
x

u (V1H )

E
x

u (V1H )
n*

E S

=
= 0 for n = 1,2,...N

u (V1H )

E
x

()
Now suppose that a risk neutral probability Q exists. We set V0H = v and then
define a utility function u as follows:

70

u ( x, ) =

x {}
P {} B1

Note that we are using the more general definition of a utility function here
(i.e. of a mapping on ). Clearly this satisfies all the properties required of
utility functions. Take arbitrary portfolio ( H 1 , H 2 ,...H N ) . Then

E u B1 ( v + H 1S 1* + H 2 S 2* + ...H N S N * ) ,

B1 ( v + H 1S 1* ( ) + H 2 S 2* ( ) + ...H N S N * ( ) ) Q {}
P {} B1

P {}

= ( v + H 1S 1* ( ) + H 2 S 2* ( ) + ...H N S N * ( ) ) Q {}

= v + E Q H n S n* = v
n =1

This applies to all trading strategies H. Thus they are all trivially optimal.
This theorem has to be interpreted correctly. It does not say that existence of
a risk neutral probability is equivalent to the existence of optimal portfolio
solution given any utility function one pleases. It says that existence of risk
neutral probability is equivalent to having some utility function for which the
optimal portfolio solution exists. In fact the utility function which we used in
our theorem is pretty useless for it makes all portfolios equally profitable!
In practice one would be more interested in finding the optimal portfolio for a
given utility function applied to a future scenario of market prices for different
assets. The direct solution to this problem is clearly a classical one in optimization.

71

Example 3.5
Suppose we use the utility function u ( x ) = 1

1
and we take the discounted
1+ x

price process captured by the table below:

t=1
2

8/7

8/7

8/7

8/7

14

16

14

21

12

21

24

22

24

30

Assets

t=0

B1S0n

Riskless

Asset 1
Asset 2

The risk neutral probability for the above problem is given by:
1
1
1

Q {
Q
Q
=
,

=
,

=
}
{
}
{
}
1
2 3 3 6
2

This probability guarantees the viability of the model, but the optimum might be
achieved with some utility function different from the one we have. For our utility
functions if we are to have an optimum the following condition is necessary:
n* u (V1H )
E S
= 0 for n = 1, 2
x

Now
u
1
1
u (V1H ( ) ) =
=
2
x
(1 + V1H ( ) ) 1 + B1 (V0H + H 1S 1* ( ) + H 2S 2* ( ) )

For n=1: S 1* (1 ) = 1.75, S 1* (2 ) = 4.375, S 1* (3 ) = 3.5


For n=2: S 2* (1 ) = 1.75, S 2* (2 ) = 0, S 2* (3 ) = 5.25

72

Then
3
S 1* (i ) P {i }
1* u
H
E S
u (V1 ) =
x

i =1 1 + B1 (V0H + H 1S 1* (i ) + H 2 S 2* (i ) )

1.75P {1}

(1 + B (V

H
0

1.75 H 1 1.75 H 2 )

4.375P {2 }

) (1 + B (V

H
0

+ 4.375 H 1 + 0 H 2 )

3.5P {3}

1 + B1 (V0H 3.5 H 1 + 5.25 H 2 )

=0

and
3
S 2* (i ) P {i }
2* u
H
E S
u (V1 ) =
x

i =1 1 + B1 (V0H + H 1S 1* (i ) + H 2 S 2* (i ) )

1.75P {1}

(1 + B (V

H
0

1.75 H 1 1.75H 2 )

0P
{2 }

) (1 + B (V
1

H
0

+ 4.375 H 1 + 0 H 2 )

5.25P {3}

(1 + B (V
1

H
0

3.5H 1 + 5.25H 2 )

=0

There are two unknowns H 1 and H 2 . Solving the two equations above in two
variables would require the use of a computer.

3.2 COMPUTING THE OPTIMUM THROUGH RISK


NEUTRALITY FOR COMPLETE MARKETS

73

The method used in the example above involves optimization over general
nonlinear derivatives of utility functions. An efficient method has been devised to
take care of this. Firstly we note that the optimum we are interested in can be
viewed as the composition of two functions: the first function maps strategies into
the random variable of their final values and the second function gives the
expectation of each such random variable. What we can do is first find the optimal
value W * = arg max E u (W ) . Here we are optimizing over all contingent claims
W

W. Since := {1 ,...K } , the set of contingent claims can be identified with K .


Having found this optimal value, we find that strategy which delivers this value.
The previous chapter has given us a method to allow us to compute a replicating
portfolio. We shall limit ourselves to complete markets so this can be done.
So we first solve the problem of finding the optimal E u (W ) . We define the set
W of random variables as follows:

W = {random variables W : R such that E Q [W / B1 ] = V0H }


So any contingent claim W is replicable by corresponding portfolio H,
V1H ( ) = W ( ) and we remember that the risk neutral valuation principle says for

any trading strategy H with initial value V0H ,


E Q [V1H / B1 ] = V0H

So we need to solve the optimization problem:


max E u (W )
W W

74

We attack this problem by the use of the Lagrange multipliers.

Indeed

max E u (W ) becomes:
W W

maxK E u (W ) E Q V0H
(W , )
B1

Q ( )
.
P ( )
ignored, so the objective function becomes:

As we defined before, let L ( ) =

Since V0H is constant this can be

L ( )W ( )

LW
E u (W )
=

u
W
,
(
)
(
)

P {}
B1
B1

W, a vector with components indexed by elements , will be the maximum


achieving wealth, when for each ,

L ( )W ( ) u (W ( ) ) L ( )

=0
u (W ( ) )
=
x
B1

x
B
1

These are
u (W ( ) )
x
any .

necessary conditions. But with


=

L ( )
B1

u
and L
x

always

positive,

for each . Otherwise the derivative cannot be zero at

These conditions can be rearranged as:


u (W ( ) )
x

L ( ) Q ( )
=
B1
B1P ( )

which gives:
u (W ( ) )
Q ( ) =

B1P ( )

75

Earlier, under optimality, we derived the equations:


u (V1H ())
Q() =

P {}
x
H

(u (V1 ))
E


(u (W ))
For each . This forces = B1E
.

This gives us a method for computing the optimal W. Denoting by I the inverse to

u
with reference to its first variable, we have:
x

L ()
u (W ())

I
= W () so W () = I
x

B1

To obtain an explicit formula for we see that for the contingent claim W,
E Q [W / B1 ] = V0H , which in our case translates to the equation:
L ()

I
B
1
= V0H (*)
E Q

B1

This equation has a unique solution because of the nature of the function I. The
expected value of W will give the optimal wealth.
Being in a complete market we can work out the corresponding portfolio H.
Piecing all together, having the discount factor and future prices scenario we
proceed as follows:
compute Q, L and I

76

then solve equation (*) to obtain and hence the optimal wealth W
find the portfolio component H = ( H 1 , H 2 ,...H N ) which generates the optimal
wealth W.
Example 3.6
Consider the following pricing scenario for 2 assets and an underlying riskless
asset:

t=1
2

10/9

10/9

10/9

St1

5.4

St2

13

0.5

0.25

0.25

Assets

t=0

Bt

Probability

The risk neutral probability is given by the solution to:

Q {1} + Q {2 } + Q {3} = 1
6Q {1} + 8Q {2 } + 4Q {3} = 6
13Q {1} + 9Q {2 } + 8Q {3} = 10
1
which is Q {1} = Q {2 } = Q {3} = . We next take the utility function:
3
u ( x ) = exp ( x )

77

Then

u (W ( ) )
x

= exp ( W ( ) ) and I ( y ) = log ( y ) . Then if we apply the

equation:
u (W ( ) )
L ( )
W ( ) = I
= I

x
B1

we get:
L ( )
L ( )
W ( ) = log
= log ( ) log

B1
B1

(**)

L ( )
I

B1

Furthermore E Q
= V0H becomes:

B1

V0H

L ( )

L
L ( )
I
log
log

B
B
log

(
)

=E
=
B1
= EQ

E
Q
Q
B1

B1
B1
B1

This gives the formula:

L

B1

H
= exp BV
1 0 E Q log

which when we plug it in (**), gives the value for the optimal portfolio W whose
utility is given by:
log ( L ( ) )
L ( )
u (W ( ) ) = exp
=

B1
B1

with expected value given by:


78

E u (W ) =

E [L]
B1

B1

(***)

We can now put numbers: The state price vector L is given by:
L(1 ) =

Q(1 ) 1/ 3 2
Q(2 ) 1/ 3 4
=
= , L(2 ) =
=
= = L(3 ) .
P (1 ) 1/ 2 3
P (2 ) 1/ 4 3

So

L 1
2
2.9
4.9
E Q log = log
+
2
log

3.10 = 0.04873
3
3.10

B1 3
Then the optimal attainable wealth is given by:

L( ))
L( )
W ( ) = log
= log( ) log
=
B
B
1

1
L
L( )
B1V0H + EQ log log

B1
B1
Thus

W (1 ) =

L
2 10 H
10 H
V0 + EQ log log
= V0 + 0.46209
9
B
3
B
1
1 9

W (2 ) =

L
2 10 H
10 H
V0 + EQ log log
= V0 0.23105 = W (3 )
9
B
3
B
1 9
1

and

10 H

V0 + 0.04873
9

= exp

The optimal value of the objective function is 10 .

79

Having computed the random variable W which gives the optimal expected wealth,
we can work backwards to obtain the corresponding trading strategy.
So we need to solve the following:
10
9

10
9

10
9

10 H

6 13
V
+
0.46209
0
9

0
H

1 10 H

8 9 H =
V0 0.23105

H 2

10
H
V0 0.23105
4 8

H 0 = V0H 1.1783, H 1 = 0.0385, H 2 = 0.1540

80

3.3 CONSUMPTION-INVESTMENT MODELS


The problem studied here involves the division of a given initial sum of money v
into a part C0 which will be consumed at t = 0, another part C1 which will be
consumed at t = 1 and a remainder which will be invested, so that the total utility of
the consumed parts will be maximized. First we need to ensure that we define the
problem properly. We assume a probability space for an underlying assets prices
process is given.

Definition 3.7
The pair (C0, C1) of a real number C0 and random variable C1 are said to consitute
a consumption process. Providing a trading strategy H to form (C0, C1, H) gives a
consumption-investment plan.
Definition 3.8
Given an initial sum of money v 0, consumption-investment plan (C0, C1, H) for
which V0H is the value of H at t = 0 and V1H at t = 1, is said to be admissable if
v = C0 + V0H and C1 = V1H

with
N

H
0

= H + H S and V1 = H B1 + H n S1n
0

n =1

n
0

n =1

81

As is customary the first question which we need to ask is whether we can find a
convenient way of checking whether a given consumption-investment plan is
admissable.
And the answer to this question is yes with the check being given by the usual riskneutral probability:
Theorem 3.9
Given initial amount v 0 and consumption process (C0,C1), trading strategy H
such that the consumption-investment plan (C0, C1, H) is admissable exists if and
C
only if for every risk-neutral probability Q we have: C0 + EQ 1 = v .
B1

Proof

()
Suppose there exists a trading strategy H such that the consumption-investment
plan (C0, C1, H) is admissable. Then v = C0 + V0H and C1 = V1H assures us that C1 is
an attainable contingent claim and we know from theorem 2.16 that
C
C
E Q 1 = V0H for any risk neutral measure thus C0 + EQ 1 = v .
B1
B1

( )
C
On the other hand if equation C0 + EQ 1 = v holds for all risk neutral
B1

probabilities then C1 is attainable and replicating portfolio H exists.

82

This theorem again signals to us the possibility for solving the consumptioninvestment problem. We can use a risk-neutral computational approach. Let us state
the problem with precision:

3.3.1 Computing Optimal Solutions


Portfolio H needs to be identified which satisfies the optimization problem:
u ( C0 ) + E u ( C1 )
max
H

subject to
N

C0 + H 0 + H n S0n = v
n =1
N

C1 H B1 H n S1n = 0
0

n =1

C1 , C0 0

Example 3.10
Given the following price process involving three assets in t=1 prices:
S0n

B1S0n

Riskless Asset

64/61

64/61

64/61

64/61

64/61

Asset 1

732

768

790

750

770

790

Asset 2

1525

1600

1580

1720

1534

1590

Asset 3

2196

2304

2180

2180

2428

2180

1/8

1/8

Probability

Consider as utility function u ( x ) =

x . Then
83

max
u ( C0 ) + E u ( C1 )
H

becomes
1
1
1
1
max C0 + E C1 = max C0 +
C1 (1 ) +
C1 (2 ) +
C1 (3 ) +
C1 (4 )
H
H
2
4
8
8

with
C0 = v H 0 732 H 1 1525 H 2 2196 H 3
and the other consumption components given by:
64 0
H1 + 790 H 1 + 1580 H 2 + 2180 H 3
61
64
C1 (2 ) = H10 + 750 H 1 + 1720 H 2 + 2180 H 3
61
64
C1 (3 ) = H10 + 770 H 1 + 1534 H 2 + 2428 H 3
61
64
C1 (4 ) = H10 + 790 H 1 + 1590 H 2 + 2180 H 3
61
C1 (1 ) =

So our problem to maximize the expression :


v H 0 732 H 1 1525H 2 2196 H 3 +
1 64 0
H + 790 H 1 + 1580 H 2 + 2180 H 3 +
2 61
1 64 0
H + 750 H 1 + 1720 H 2 + 2180 H 3 +
4 61
1 64 0
H + 770 H 1 + 1534 H 2 + 2428H 3 +
8 61
1 64 0
H + 790 H 1 + 1590 H 2 + 2180 H 3
8 61

84

This is an expression in the portfolio components and so to obtain the optimal


0
1
2
3
solution in terms of them we differentiate with respect H , H , H ,H and equate to

0 to obtain 4 equations:

1
2 v H 0 732H1 1525H 2 2196H 3

64 / 61

64 0
H + 790H 1 +1580H 2 + 2180H 3
61
64 / 61
64 / 61
+
64 0
64 0
8
H + 750H 1 +1720H 2 + 2180H 3 16
H + 770H 1 +1534H 2 + 2428H 3
61
61
64 / 61
+
=0
64 0
16
H + 790H1 +1590H 2 + 2180H 3
61

732
2 v H 0 732 H 1 1525H 2 2196H 3

790

64 0
H + 790 H 1 + 1580 H 2 + 2180 H 3
61
750
770
+
64 0
64 0
8
H + 750 H 1 + 1720 H 2 + 2180 H 3 16
H + 770 H 1 + 1534 H 2 + 2428H 3
61
61
790
+
=0
64 0
16
H + 790 H 1 + 1590 H 2 + 2180 H 3
61
4

85

1525
2 v H 0 732 H 1 1525H 2 2196H 3

1585

64 0
H + 790 H 1 + 1580 H 2 + 2180 H 3
61
1720
1534
+
64 0
64 0
8
H + 750 H 1 + 1720 H 2 + 2180 H 3 8
H + 770 H 1 + 1534 H 2 + 2428H 3
61
61
1590
+
=0
64 0
1
2
3
16
H + 790 H + 1590 H + 2180 H
61
2196
2180
+
2 v H 0 732 H 1 1525H 2 2196H 3 4 64 H 0 + 790 H 1 + 1580 H 2 + 2180 H 3
61
2180
2428
+
64 0
64 0
H + 750 H 1 + 1720 H 2 + 2180 H 3 8
H + 770 H 1 + 1534 H12 + 2428H 3
8
61
61
2180
+
=0
64 0
1
2
3
H + 790 H + 1590 H + 2180 H
16
61
4

And we solve the above nonlinear equations.


Generally we have to solve simultaneously for :

C
C
u
u
u
u
(C0 , ) 00 + E[ (C1 ) 10 ] = (C0 , ) + B1E[ (C1 )] = 0 ...(A)
x
x
x
x
H
H

C
C
u
u
u
u
(C0 , ) 0n + E[ (C1 ) 1n ] = S0n
(C0 , ) + E[ (C1 ) S1n ] = 0 ...(B)
x
H
x
H
x
x
for n =1,2,, N.

If we write the equations above as F(H) = 0, then we can iterate the NewtonRaphson way:
86

H( k +1) = H( k ) J (H( k ) )1 F (H( k ) )


And in general this would be done through constrained optimization methods.
However there is one point we made no comment about. Are we sure that the
solution does not involve negative values either in constant C0 or in random variable
C1?
Well these conditions should be heeded. This may be imposed via the choice of
appropriate utility functions like u(x)=log(x).
But as usual we do not content ourselves with using nonlinear maximization
methods blindly. We look for solutions which are easier to obtain if we cleverly use
the risk neutral probability. In fact from (A) above we see that

u
u
(C0 ) = B1E[ (C1 )]
x
x
u

x (C1 ) B1
1= E

u (C0 )
x

Letting

u
(C1 ) B1
L := x
u
(C0 )
x

suggests to us to put :

u
(C1 ( ))

x
Q { } = B1
P { }
u
(C 0 )
x
We can see that this implies the equation:

87

u
( C1 ( ))
n
n

x
E Q S1 = S1 ( ) B1
P { }
u

(C 0 )
x
B1
u

=
E
(C1 ) S1n = B1 S 0n
u
x

(C 0 )
x
where the last equality follows from (B) above. Thus Q is indeed a risk neutral
probability measure.

3.3.2

Optimal Solution Constructed through Risk Neutral


Probability

The reasoning above gives us the basis for the risk neutral oriented solution to the
consumption-investment problem. The idea is to treat C1 as a contingent claim
which should be the solution to:

max u (C0 ) + E[u (C1 )]


C0 , C1

subject to
C
C0 + E Q 1 = v
B1
C0 , C1 0
Using the Lagrange multiplier technique, this problem then translates to :

88

max u (C0 ) + E[u (C1 )] {C0 + E Q [C1 / B1 ] v} =


C0 , C1

max u (C0 ) + E[u (C1 )] {C0 + E[C1L / B1 ] v}


C0 , C1

Q { }
L ( ) =
where
P { } .
We shall assume that the choice of utility function will automatically rule out
negative consumption values. Then we can set up the set of equations which give us
the optimum:

L ( )
u
u
(C0 ) = and
(C1 ( )) =
x
x
B1
Then if we define the function:
1

u
I ( x) = ( x )
x
we have

C0 = I ( )
and

C1 ( ) = I ( L( ) / B1 ) .
Then to solve for we use

C0 + E Q [C1 / B1 ] = v = I ( ) + E Q [ I ( L / B1 ) / B1 ] .

89

Example 3.11 (Continuation of example 3.10).


1

I ( x) = ( u ' ) ( x ) =

1
4 x2

gives us:

B12
v = I ( ) + E Q [ I ( L / B1 ) / B1 ] = 2 + E Q 2 2
4
4 L B1
1

and = 2 C allows us to substitute for :


0
B
B
E Q 21 2 = 12
4 L 4

{i }

i =1

L(i ) 2

64C0
=
61

Q {i }

i =1

L(i ) 2

64C0
=
61

P {i }

Q { }
i =1

and
2

4 Q { }
4 P { }
B12
64
64
i
i

v=
+ E Q 2 2 = C0 +
C0
= C0 1 +

2
2

4
61 i =1 L(i )
61 i =1 Q {i }
4 L B1

So first we work out the risk-neutral probability:


This turns out to be
Q {1} = 0.1, Q {2 } = 0.3, Q {3 } = 0.5, Q {4 } = 0.1

Also
P {1 } = 0.5, P {2 } = 0.25, P {3 } = 0.125, P {4 } = 0.125

so that we compute
L (1 ) = 0.2, L (2 ) = 1.2, L (3 ) = 4, L (4 ) = 0.8

90

Then from above we compute:

v = 4.0383C0
which gives
C0 = v / 4.0383 = 0.2476v

so that

1
2 C0

1.0096
v

with future consumption given by:

0.2726v
1 B1 1
64 v
C1 () = I (L() / B1 ) =
=

4 L() 4 61 1.0096L()
L()2

from equations above. So

C1 (1 ) = 6.8145v
C1 (2 ) = 0.1893v
C1 (3 ) = 0.0170v
C1 (4 ) = 0.4259v
and we have obtained C1 as a contingent claim now.
We could work out the replicating portfolio by solving for H in :

91

B1 H 0 + H 1 S11 (1 ) + H 2 S12 (1 ) + H 3 S13 (1 ) = 6.8145v


B1 H 0 + H 1 S11 (2 ) + H 2 S12 (2 ) + H 3 S13 (2 ) = 0.1893v
B1 H 0 + H 1 S11 (3 ) + H 2 S12 (3 ) + H 3 S13 (3 ) = 0.0170v
B1 H 0 + H 1 S11 (4 ) + H 2 S12 (4 ) + H 3 S13 (4 ) = 0.4259v

This gives us the full consumption-investment plan (C0,C1,H) and thus our problem
has been conveniently split up into two parts.

3.4 MEAN-VARIANCE PORTFOLIO PROBLEM


This classical problem was first studied by Harry Markowitz in his paper Markowitz,
Harry M. (1952). "Portfolio Selection". Journal of Finance 7 (1): 77-91 a line of
work within mathematical finance which earned him a Nobel prize in Economics in
1990.
T

Let us first take care of the context. Given portfolio H = ( H 0 , H 1 , H 2 ,...H N ) with
H

initial value V0H > 0 , its return RV is given by


VH

V1H V0H
=
V0H

Earlier we saw that we could split up the return of a portfolio into the sum of returns
from each asset:
H

RV =

N
H0
H n S0n n
r
+
R

H
V0H
V
n =1
0

This shows that we can write any return of portfolio as:

92

RV = (1 F1 ...FN ) r + Fn R n (*)
H

n =1

where Fn is interpreted as the fraction from the total wealth possessed at t = 0 which
is assumed by the nth risky asset (i.e. Fn =

H n S0 n
) . Thus we can write the average
V0 H

return of portfolio as:


N

E RV = (1 F1 ...FN ) r + FnE R n

n =1
H

Clearly any rational investor (an investor who behaves rationally in the sense that he
or she always prefers more to less) chooses a portfolio H with the aim of obtaining a
return that is as large as possible. If the average return is the only criterion for this
judgment then clearly one will invest the whole money into that asset which gives the
highest mean return. However, this asset may be very risky and thus the returns can
have huge fluctuations. The idea of Markowitz was to find a tradeoff between the
risk incurred from the portfolio and the returns of the same portfolio. He assumed
that the risk of the portfolio is measured by the variance of the returns and thus
considered two optimization problems:
1. Obtaining a portfolio which minimizes the risk given a fixed (lower bound)
mean return
2. Obtaining a portfolio which maximizes the mean return for a given level of
risk (upper bound).
In this study unit we shall focus only on the first problem. A compact formulation of
this problem is the following:

93

Problem 3.12

Given
a deterministic interest rate r
a fixed with r
that there exists one portfolio with return not equal to r
determine portfolio H* with return RV

H*

= R such that

R = arg min Var RV

H
H

(A)

subject to
E RV =

Using (*) the mean-variance portfolio problem can be reformulated equivalently as a


quadratic problem:
R = arg min F ' C
F
F

(B)

subject to
N

(1 F1 ...FN ) r + Fn E R n =
n =1

where C is the N N covariance matrix for the returns of the assets:

C ij = Cov[ Ri , R j ]
T

and F = ( F1 ,...FN ) . This is a standard problem in optimization and its solution is


pretty classical, offering an attractive alternative way for solution.
Another way is through the risk neutral computational approach. To implement this
we need yet another formulation which is given below:

94

V1H * = arg min Var V1H


H

subject to
E V1 = v (1 + )
V0H = v
H

(C)

This will provide us with a half-way measure towards getting a formulation which
will enable us to somehow compute solutions numerically with a certain ease.

Theorem 3.13

There is a 1-1 correspondence between solutions to problem (A) and those to


problem (C).
Proof:

First we show that the constraints give the same solution sets.
H

RV V0H = V1H V0H


H

E[ RV ]V0H = E[V1H ] V0H


H

E[V1H ] = V0H (1 + E[ RV ])
so that both conditions are telling us that
H

E[V1H ] = v (1+ ) and V0H = v E[ RV ]=


H
Let V1 be a solution of (C). Then its corresponding return

V1 H V0H
R=
V0H

should satisfy constraints above and have minimal variance:


If V1H satisfies constraints then its return has minimal variance:

95

Var R =

Var V1 H

(V )
H
0

Var V1H

(V )
H
0

= Var RV

On the other hand if R is a solution to (A) then


Var (1 + R ) v = v 2 Var R

v 2 Var RV

V1H v
2
= v Var

v
H

= Var V1H

and so Var (1 + R ) v is a solution to (C).

Another equivalent formulation which will be used to solve the Mean-Variance


Portfolio problem through the risk neutral computational approach is given below
2
1

V1H * = arg max E (V1H ) + V1H


H
2

subject to
(D)

V0H = v

We give the condition on which establishes equivalence:

Theorem 3.14

If Q is a risk neutral probability for which equation

V0H ( (1 + )E Q [L] B1 )
E Q [L] 1

96

holds, then Problems (B) and (D) are equivalent.

Proof:
H
Suppose (B) gives us as solution V1 .
2

Then Var V1 H = Var V0H R V = (V0 H ) ( F' CF ) is a minimum and also for any

other feasible V1

we have E R

VH

V1H V0H
= = E
, which shows that
H

V
0

E V1H = V0H (1 + ) . Thus


2
1
1
1

E V1H
V1H = E V1H Var[V1H ] E V1H
2
2
2

2
1
1
V0H (1 + ) V0H (1 + ) Var[V1H ]
2
2
1

1
V0H (1 + ) V0H (1 + ) Var[V1 H ] =
2

( )

1
1

V1 H
E V1H E V1 H Var[V1 H ] = E V1H
2
2

( )
2

so that (D) is indeed solved.


Now suppose we have a solution to (D). Then if we look at the objective function as
though it were a utility function

u ( x) = x

1 2
x
2

97

u
then x ( x) = x and I ( x) = x then by previous formulas in section 3.2 the
solution for the optimal wealth which we denote by V1

should satisfy the following

two equations:

L( ))
L( ))
V1 H ( ) = I
=
B1
B1

B1 ( B1V0H )
L
H
E Q
B1 = V0 =
B
EQ [L]

1
giving

( B1V0H ) L( ) ( E Q [ L ] L( ) ) B1V0H L( )
V1 ( ) =
=
+
EQ [ L]
EQ [ L]
EQ [ L]
H

which in turn gives :

E V1 =
H

( E Q [ L ] 1)
EQ [ L]

B1V0H
+
EQ [ L]

and we need to have

B1V0H
E V1 = V (1 + ) =
( EQ [ L] 1) + E [ L]
EQ [ L]
Q
H

H
0

This implies that

( E Q [ L ] 1) = V0H (1 + )E Q [ L ] B1V0H
and thus

V0H ( (1 + )E Q [L] B1 )
E Q [L] 1

98

The formulations above allow us to give a number of results which have important
bearing on some classical problems of mathematical finance.

Theorem 3.15

The return R of the portfolio corresponding to the optimal solution to the meanvariance portfolio problem is an affine function of state price density L :

R ( ) =

E Q [L] r

L ( )
E Q [L] 1 E Q [L] 1

Proof:

This follows from the previous theorem where we got :

( B1V0H )
L( ) = a + bL( )
V1 ( ) =
EQ [ L]
H

noting that is not dependent on .


Furthermore equivalence of problems (B) and (D) allows us to ascribe the properties
of solutions to (D) to solutions to (B).
Then

V1 H ( ) V0H
1
R ( ) =
= H
H
V0
V0

V0H

( B1V0H )

L
(
)

1 =
E
L
[
]
Q

( B1V0H )
1 H
L( ) =
V E [ L ]
0
Q

(1 + )E Q [L] B1
E Q [L] 1

(1 + )E Q [L] B1
B1
1

L( ) =
(E [L] 1)E [L] E [ L ]
Q
Q
Q

99

E Q [L] ( B1 1) (1 + )E Q [L] B1E Q [L]


E Q [L] 1

L( ) =

(E Q [L] 1) E Q [ L]

E Q [L] ( B1 1) ( B1 1)

E [L] 1
Q
E Q [L] r r

L( )
E Q [ L] 1 E Q [L] 1
E Q [L] 1

L( ) =

We next give a classical result of a certain importance in mathematical finance which


uses a result we proved at the end of last chapter.

Theorem 3.16 (CAPM)

If R is the return of any given portfolio and R the return of the optimal solution to
the mean-variance portfolio problem then if r we have:

E[ R ] r =

Cov[ R, R ]
E[ R ] r
V
ar[ R ]

Proof :

To prove this result we can immediately apply Theorem 2.26, in the previous
chapter, to R .

Example 3.17

100

The table below gives the price scenarios for 4 risky assets and the rate for the
riskless asset. We would like to work out the solution to the corresponding meanvariance problem.

S1n

S0 n

Assets

Riskless

576/540

576/540

576/540

576/540

576/540

990

1032

1066

1066

1050

1038

1980

2132

2109

2132

2096

2096

3960

4242

4219

4222

4244

4208

5940

6349

6326

6336

6366

6324

0.25

0.25

0.1

0.2

0.2

Probability

r is given by 576/540 1 = 36/540 = 0.0667


The risk-neutral probability works out at :

Q (1 ) = 1/10, Q (2 ) = 2 / 5, Q (3 ) = 1/ 5, Q (4 ) = 0.15, Q (5 ) = 0.15


The returns matrix is given by:

1
Riskless

0.066667 0.066667 0.066667 0.066667 0.066667

0.04242

0.07677

0.07677

0.06061

0.04848

0.07677

0.06515

0.07677

0.05859

0.05859

0.07121

0.06540

0.06616

0.07172

0.06263

0.06886

0.06498

0.06667

0.07172

0.06465

101

Probability

0.25

0.25

0.1

0.2

0.2

The corresponding mean returns are given by

E [ R1 ] =0.0593
E [ R2 ] =0.0666
E [ R3 ] =0.0676
E [ R4 ] =0.0674
and the corresponding returns covariance matrix C is:
RA

RB

RC

RD

RA

0.0002023

-0.000015

-0.0000136 -0.0000106

RB

-0.000015

0.0000652

0.0000133

0.0000032

RC

-0.0000136 0.0000133

0.0000171

0.0000115

RD

-0.0000106 0.0000032

0.0000115

0.000008

So the problem is given by:


Find positive fractions F1, F2, F3 , F4 such that for a given :

(0.0593-0.05)F1 + (0.0666-0.05)F2 +
(0.0676-0.05)F3 + (0.0674-0.05)F4 = 0.05
holds and the quadratic form:

( F1

F2

F3

0.2023

-0.015
F4 )
-0.0136

-0.0106

-0.0136 -0.0106 F1

0.0652 0.0133 0.0032 F2
0.0133 0.0171 0.0115 F3

F4
0.0032 0.0115 0.008
-0.015

102

is minimized.
We have mulitplied all entries in the matrix by 1000 to make the matrix more
presentable. And we can use the standard methods of quadratic programming.

Let us next solve the same problem using the methods given Theorem 3.14.

L( )=

Q( )
P ( ) gives

0.4

0.75

0.75

and

E Q [ L ] = 1.305
Then

V0H ( (1 + )E Q [L] B1 )
E Q [L] 1

V0H (1.305(1 + ) 576 / 540 )


0.305

4.27869V0H (1 + ) 3.49727
and then

103

V1 ( ) =

( E Q [ L ] L( ) )
EQ [ L]

B1V0H L( )
+
=
EQ [ L]

4.27869V0H (1 + ) 3.49727 (1.305 L( ) )


1.305
( B1V0H )L( )
H
=
V1 ( ) =
EQ [ L]

576V0H L( )
+
=
1.305 540

= 4.27869V0H (1 + ) 3.49727
(4.27869V0H (1 + ) 3.49727 576V0H / 540)L( )
=
1.305
4.27869V0H (1 + ) 3.49727

( 3.278689V

H
0

(1 + )-2.679899-0.81737V0H ) L( )

104

Chapter 4
Multi-period Discrete Financial Markets
The material presented in the previous chapters shall now be generalized to a
multi-period discrete setting. Apart from few concepts that were not relevant in a
single period environment the various results presented earlier can be extended in a
straight forward manner to multi-period financial markets.

4.1. THE STOCHASTIC BASE FOR MULTI-PERIOD MODELS


We shall start by providing the necessary mathematical framework
= {0,1, 2,...T } ,
Time set : T

Set of all possible outcomes :

= (1 ,...T ) : t { K tmin , K tmin 1,... 1, 0,1,...K tmax 1, K tmax }

where Ktmax {0} , Ktmin for all t, on which a non-zero probability is


defined
-algebra : FP
= () ,
Filtration :
o F0
= {, } (trivial -algebra) ,

105

} ,

( K1min , K 2min ,... KTmin ) , ( K1min , K 2min 1,... KTmin ) ,... ( K1min , K 2max ,...KTmax )

min
min
min
min
min
min

( K1 1) , K 2 ,... KT , ( K1 1) , K 2 1,... KT ,

,
F1
= ... ( K min 1) , K max ,...K max

T
1
2

...

K max , K 2min ,... KTmin ) , ( K1max , K 2min 1,... KTmin ) ,... ( K1max , K 2max ,...KTmax )
( 1

)(

( K 1min , K 2m in , K 3min ... K Tm in ) , ( K 1min , K 2m in , K 3min 1, ... K Tm in ) ,

... ( K min , K m in , K max , ... K m ax )

1
2
3
T

K min , K m in 1, K min ... K min , K m in , K min 1, K m in 1, ... K min ,


(
)
(
)
1
2
3
T
1
2
3
T

F2
= ... ( K 1min , K 2m in 1, K 3m ax , ... K Tmax )
,

...

( K 1m ax , K 2m ax , K 3m in ... K Tmin ) , ( K 1max , K 2max , K 3min 1, ... K Tm in ) ,

... ( K 1m ax , K 2m ax , K 3m ax , ... K Tmin )

o Ft is the -algebra generated by the collection of sets containing

sequences that have the same elements in position one up to position t


and all possibilities in the remaining entries. This indicates that as time
progresses the values of the process will be known up to the time instant
where we would be at the moment. Thus in Ft we must make sure that
paths whose evolution up to time t are not distinguishable from one
another.
Probability measure can be set by letting

min

max

pt Kt +1 ... ptKt +1 denote the

probabilities as shown in the figure below:

106

K tmax
+1

Ktmax
+1
t

max

ptKt +1

K tmax
+1 1

.
.
j
.

min

pt K. t +1

K tmin
+1
t+1

If independence is assumed the probability of each path can be computed as


follows:
p0j1 p1j2 ... pTjT1

where each jt { K tmin ,...Ktmax } , t = 1,2,...T .

4.2. PRICE PROCESS


The financial market is made up of:
T

A Riskless Asset with unit price process { Bt }t =0 , where B0 = 1 and for each
t = 1, 2,...T , Bt = (1 + r ) Bt 1
T

N Risky Assets, each with unit price process {Stn }

t =0

For each n, {Stn }

t =0

where n = 1,2,N.
T

must be adapted with respect to the filtration {Ft }t =0


,

that is each random variable Stn must be measurable with respect to Ft .

107

As in single-period models St = ( Bt , St1 ,...StN ) where t T will denote the price


vector.

Example 4.1

Suppose N = 1, T = 2, K1min = K1max = 1 and K 2min = K 2max = 2 . Possible values of St1


which make the process adapted to with respect to

{Ft }t =0 is shown in the table

below :
St1 ( )

t=0

t=1

t=2

(1, 2 )

(1,1)

2.3

(1,0 )

2.1

(1, 1)

2.5

(1, 2 )

1.9

( 0, 2 )

2.5

2.1

( 0,1)

2.5

2.2

( 0,0 )

2.5

2.2

( 0, 1)

2.5

( 0, 2 )

2.5

3.4

( 1, 2 )

4.1

( 1,1)

4.1

( 1,0 )

4.1

6.3

108

( 1, 1)

4.1

3.8

( 1, 2 )

4.1

3.6

4.3. DISCOUNTED PRICE PROCESS AND RETURN PROCESS


Associated to each price process, there is the discounted price process and the
return process.

Definition 4.2
T

For each n, the discounted price process {Stn*}

t =0

is defined by :

Stn
=
Bt

n*
t

Definition 4.3
T

The return process { Rtn } is defined as follows:


t =0

R0 ( ) = 0
Stn ( ) Stn1 ( )
for t = 1, 2,...T
Stn1 ( )

Rtn ( ) Rtn1 ( ) := Rtn ( ) =


n

Note that Rt ( ) 1 due to the non-negativity of the asset prices. The


following equations also hold:
t

St = S 0 + Su 1Ru
u =1
t

St = S0 (1 + Ru )
u =1

109

4.4. BINOMIAL MODEL


In general multi-period models, at each time instant, the price of an asset would be
open to a number of changes in value subject to a probability distribution.
In the binomial model at each time t the process has two options open. It can either
go up or down. In this case, we set the stochastic base
= {(1 ,...T ) : t {1,0}}

(i.e. Ktmin = 1 and K tmax = 0 for all t T )


In this model:
F0 is the trivial -algebra.
F1 is the -algebra generated by the partition of with the two subsets : the one

whose sequences start with 0, the other whose sequences start with 1.
Ft is the -algebra generated by the partition of with the 2t subsets each subset

containing sequences start with some particular pattern of t 0's and 1's.

The probability measure on this measurable base can now be set up by knowing
the probability p of an upward move and then a downward move happens with
probability 1-p. In the simplest set-up we would have independent moves with the
same probability and we find the probability of each path of finite length by
multiplying the probabilities of all the individual moves.
Let us define the process Nt, which tells us at time t how many moves upwards we
have. So

110

N t ((1 ,.., i ,.., K )) = ( j + 1)


j =1

and

P[{ : Nt ( ) = k }] = tCk p k (1 p)t k

for k = 0,1,..., t

The prices of the risky asset evolve as follows:

{ : St +1 ( ) = uSt } = p
P
{ : St +1 ( ) = dSt } = 1 p
P
So given the initial price S0

P[{ : St ( ) = S0u k d t k }] = t Ck p k (1 p)t k

for k = 0,1,..., t

Thus
t

E[ St ] = t Ck p k (1 p )t k S 0u k d t k
k =0

1
2

= S0 t Ck (up ) k (d (1 p ))t k
k =0

= S0 (up + d (1 p ))t
The returns process is given by :

St St 1 S0u Nt ( ) d t Nt ( ) S0u Nt 1 ( ) d t 1 Nt 1 ( )
Rt =
=
St 1
S0u Nt 1 ( ) d t 1 Nt 1 ( )
= u Nt ( ) Nt 1 ( ) d 1 ( Nt ( ) Nt 1 ( )) 1
so that

111

u 1 with probability p
Rt =
d 1 with probability 1-p

4.5. TRINOMIAL MODEL


In the trinomial model at each time t the process has three options open. It can
either go up, down or stay level. In this case, we set the stochastic base
= {(1 ,...T ) : t {1,0,1}}

(i.e. Ktmin = 1 and K tmax = 1 for all t T )


In this model:
F0 is the trivial -algebra.
F1 is the -algebra generated by the partition of with the three subsets : the one

whose sequences start with 1, another whose sequences start with 0 and the other
whose sequences start with 1.
Ft is the -algebra generated by the partition of with the 3t subsets each subset

containing sequences start with some particular pattern of t 1s, 0's and 1's.

The probability measure on this measurable base can now be set up by knowing
the probability p1 of an upward move, p2 probability of staying level and then a
downward move happens with probability 1-p1-p2.
The prices of the risky asset thus evolve as follows:

112

So given the initial price S0:

P[{ : St () = S0u k d l mt k l }] =

t!
t -k -l
p1k p2l (1 p1 p2 )
k !l !( t k l )!

for k , l = 0,1,..., t , k + l t

where m is the level factor (used to help us visualize the lattice better). Thus the
expectation of St is given by
t

t k

E[ St ] = S0
k =0 l =0
k

( t k )! p d l (1 p p )m t k l
( 2 )(
)

1
2
!

!
!

!
k
t
k
l
t
k
l
( ) l =0 (
)
k =0
k
t
t !( p1u )
t k
=
( p2 d + (1 p1 p2 )m )
k = 0 k !( t k ) !
t
= ( p1u + p2 d + (1 p1 p2 ) m )
t

= S0

t !( p1u )

t!
p1k p2l (1 p1 p2 )t k l u k d l mt k l
k !l !( t k l )!
t k

where the last equality follows from the trinomial expansion:


n

( x + y + z ) = ( x + ( y + z ) ) = n Ck x k ( y + z )
n

nk

k =0
n nk

n nk

k =0 l =0

k =0 l =0

= n Ck n k Cl x k y l z n k l =

n!
x k y l z n k l
k !l !( n k l ) !

113

Letting Nt be the random variable corresponding to the number of upward moves in


t trials and Mt the random variable corresponding to the number of downward
moves, the returns process may be written as :
N ( ) N t 1 ( ) + M t ( ) M t 1 ( ) )
Rt ( ) = u Nt ( ) Nt 1 ( ) d M t ( ) M t 1 ( )) m ( t
1

so that

u 1 with probability p1

Rt = d 1 with probability p2
m 1 with probability 1 p p
1
2

4.6. TRADING STRATEGIES


In a multi-period setting, the portfolio Ht represents the number of units to carry
T

forward from time t-1 to time t. The collection H := { H t } of portfolios will again
t =1

be referred to as a trading strategy

Definition 4.4

A trading strategy H is a collection of

{( H

0
t

,...H tN )

t =1

N + 1 stochastic processes

which represents the number of units an investor carries forward

from time t 1 to time t. Each asset has its own stochastic process which tells us
how many units of the particular asset are contained within the portfolio time step
by time step for each different possible future scenario.

With a finite probability space we can visualize H as an kNT matrix where


k is the number of points in , N is the number of risky assets and T is the

114

number of time steps. Thus for fixed asset n and fixed ,

the vector

H tn ( ) : t = 1,..., T gives the investment time profile in units of the n'th asset
n
for one possible scenario. The vector H t (i ) : i = 1,.., K gives the portfolio

spread in units of the n'th asset as we scan over all possibilities. The vector

H tn (i ) : n = 1,.., N gives the portfolio spread in units over all the assets for a
particular point in time and a particular possible scenario.

Since Ht represents the number of units to carry forward from time t -1 to time t,
then the decision on how much to invest in each asset must depend on the price of
. This
the asset at time t 1. Hence Ht must be measurable with respect to Ft 1
leads to the notion of predictability of a stochastic process.
Definition 4.5
T

A stochastic process { X t }t =0
is said to be predictable with respect to the filtration
T

if for each t = 1,2,T,


{Ft }t =0

X t is Ft 1
measurable.

Definition 4.6

Suppose N = 2, T = 2, K1min = K1max = 1 and K 2min = K 2max = 2 . Possible values of

(H

0
t

, H t1 , H t2 ) which make the process predictable to with respect to

{Ft }t =0 are

shown in the table below :

Trading Strategies

(H

0
t

, H t1 , H t2 )

t=1

(H

0
t

, H t1 , H t2 )

t=2

115

4.7

(1, 2 )

( 50,20,15 )

( 40,1, 2 )

(1,1)

( 50,20,15 )

( 40,1, 2 )

(1,0 )

( 50,20,15 )

( 40,1, 2 )

(1, 1)

( 50,20,15 )

( 40,1, 2 )

(1, 2 )

( 50,20,15 )

( 40,1, 2 )

( 0, 2 )

( 50,20,15 )

( 4,3,19 )

( 0,1)

( 50,20,15 )

( 4,3,19 )

( 0,0 )

( 50,20,15 )

( 4,3,19 )

( 0, 1)

( 50,20,15 )

( 4,3,19 )

( 0, 2 )

( 50,20,15 )

( 4,3,19 )

( 1, 2 )

( 50,20,15 )

( 35,31,2 )

( 1,1)

( 50,20,15 )

( 35,31,2 )

( 1,0 )

( 50,20,15 )

( 35,31,2 )

( 1, 1)

( 50,20,15 )

( 35,31,2 )

( 1, 2 )

( 50,20,15 )

( 35,31,2 )

VALUE OF PORTFOLIO, GAINS PROCESS AND SELF -FINANCING


TRADING STRATEGIES

Definition 4.7
T

Given a trading strategy H, the corresponding value process {Vt H }


is given by:
t =0

V0H = H10 B0 + H1n S0n


n =1

116

Vt

( ) = H ( ) Bt + H tn ( ) Stn ( )
0
t

for t = 1, 2,...T

n =1

Definition 4.8
T

The gains process {GtH }


is given by
t =1

) = H (Bs Bs 1 ) + H sn (Ssn ( ) Ssn1()) t = 1, 2,...T


0
s

H
(
t

s =1

n =1 s =1

Definition 4.9

The discounted value process


H* T
t
t =0

{G }

H* T

{V }
t

t =0

and discounted gains process

are defined as follows :


N
H*
0

= H + H 1n S 0n *
0
1

n =1

Vt

H*

( )

= H () + H tn ( )Stn * ( )
0
t

for t 1

n =1

) = H sn ( )Ssn * ( )

H*
(
t

for t 1

n =1 s =1

To simplify matters, the value of the portfolio before and after transactions with
assets is retained, that is no cash is assumed to be injected into the process from
outside. This leads to the notion of self-financing trading strategy:
Definition 4.10

A trading strategy is said to be self-financing if


N

Vt H ( ) = H t0+1 ( ) Bt + H tn+1 ( ) Stn ( )

for t = 1,.., T 1

n =1

117

4.8

ARBITRAGE AND RISK NEUTRAL PROBABILITY MEASURES

The results which we obtained for a single period all generalize to multiperiod.
Exclusion of arbitrage is also equivalent to the existence of a risk neutral
probability. However risk neutral probability was then defined in terms of one set
of random variables, one set for each asset. Here each time instant has its set of
random variables! This notion of probability has to be suitably modified to fit in
with the multiperiod context:
Definition 4.11

An arbitrage opportunity in a multiperiod securities market consists of a trading


strategy H such that:
i.

H is self-financing

ii.

V0 = 0

iii.

VT() 0

iv.

E[VT()] > 0

Note that the last condition is equivalent to the fact that P VT ( ) > 0 > 0 , that is
the final wealth is positive with a strict positive probability.
Definition 4.12

A risk neutral probability measure Q on is called a martingale measure if it


satisfies the conditions:
i.
Q [{}] > 0
ii.

n*
The discounted price process St is a martingale under Q , that is for all

ts

E Q [ Stn * | Fs ] = S sn *
118

Our probability space is assumed finite and letting F s

( )

= S sn

be the sigma algebra

n
generated by the asset price at time s, the conditional expectation E Q [ St * | Fs ] is

a random variable defined by


E Q [ Stn * | Fs ] = E Q [ Stn * | : S sn * ( ) = sm ] = zQ[ Stn * = z | : S sn * ( ) = sm ]

for m = 1, 2,...card K smin , K smin 1,... 1, 0,1,...K smax 1, K smax

}
}

Example 4.13

Consider the following financial market with one risky asset and a bank account.
The price profile of the risky asset is given by:

t=0

t=1

t=2
14

12
10
11

10
8

5
i.e. N = 1 and T = 2 . Our probability space has 4 points 1= (1,1), 2 = (1,0), 3=
(0,1), 4=(0,0) (i.e. K1max = K 2 max = 1, K1min = K1min = 0 ). Suppose the bank account
rate is given by r.
n
n
We need a probability measure Q such that E Q [ S t * | F s ] = S s * for s t. Taking

any three of the following equations:

119

12 ( Q (1 ) + Q (2 ) ) + 8 ( Q (3 ) + Q (4 ) )
1+ r

14Q (1 ) + 10Q (2 )
Q (1 ) + Q (2 )

(1 + r )

= 10 for s = 0, t = 1

12
1+ r

and
11Q (3 ) + 5Q (4 )
Q (3 ) + Q (4 )

(1 + r )

8
for s = 1, t = 2
1+ r

14
Q (1 ) + 10Q (2 ) + 11Q (3 ) + 5Q (4 )

(1 + r )

= 10 for s = 0, t = 2

together with the constraint:

Q (1 ) + Q (2 ) + Q (3 ) + Q (4 ) = 1
we find that
1 + 5r 1 + 6r
Q ( 1 ) =

2 2

1 + 5r 1 6r
Q( 2 ) =

2 2

1 5r 8r + 3
Q(3 ) =

2 6

1 5r 8 r 3
Q(4 ) =

2 12

120

We shall next state the results, without proof, that generalize from single period to
multiperiod discrete models. This shows how the theory of complete markets can
be generalized in the multiperiod setting by using the martingale measure instead
of the risk-neutral probability.

Theorem 4.14

Arbitrage opportunities are excluded in a mulitperiod securities model market if


and only if there exists a martingale measure Q .

Theorem 4.15

If Q is a martingale measure and Ht a self-financing strategy, then the


corresponding discounted value process Vt

H*

is a Q -martingale.

Theorem 4.16

If there exists no arbitrage opportunity within a multiperiod model, then at no


single period submodel do we have an arbitrage opportunity.
Theorem 4.17
*

A strictly positive measure Q is a martingale measure if and only if Rt is a


martingale under measure Q .

In practice these theorems are used, just as we saw for the single period case, to
price various contingent claims. These ideas have also been utilized in more
complex situations related to optimal portfolio problems, consumption-investment
problems and others.

121

We shall finish off by going back to the binomial model which in its simplicity
reveals so many concepts with a clarity that allows deep understanding and wide
generalizability.
If we have a constant interest rate r, the martingale measure will operate
multiplicatively with common upward and downward probabilities. Thus we
proceed to calculate using :

Rtn+1 Rt0+1
E
Ft = 0
0
1
+

R
t +1

which implies that

q ( u 1 r ) + (1 q )( d 1 r )
=0
1+ r
so that q =

1+ r d
ud

Existence of martingale measure is then guaranteed by 0 < q < 1 so that u > d and u
>1+r
[{ } ] = q N t ( ) (1 q ) t N t ( )
Then Q

If we go back to example 4.13 we see that there we had different jumps for each
period. So we had q and then

Q(1 ) =

(1 + 5r )(1 + 6r )
= q2
4

122

Q(1 ) =
Q ( 2 ) =

(1 + 5r )(1 + 6r )
= q2
4

(1 + 5 r )(1 6 r )
= q (1 q )
4

Q(3 ) =

(1 5r)(1+ 4r)
= (1 q) q
4

Q(4 ) =

(1 5r )(1 4r )
2
= (1 q )
4

4.7. OPTIMAL PORTFOLIOS IN A MULTIPERIOD SETTING


In this section we generalize the optimal portfolio problem introduced in chapter
two, to a multiperiod setting.

Suppose that the initial value of the portfolio is v and H is the set of all selffinancing trading strategies with
N

v = H B0 + H1n S0n
0
1

n =1
max
We want to find that strategy H H whose expected utility at t = T is given

by:

E u VTH

max

) = max E u (V )
H H

H
T

(1)
Existence or otherwise of a portfolio whose expected utility is given as above is
called the optimal portfolio problem.

123

H
H*
Recalling that VT = BT (v + GT ) (since trading strategies are assumed to be self-

financing)

the optimization program (1) can be written as an unconstrained

program by considering for the time being the number of units to invest in the
risky assets only:

E u VTH

max

) = max E u (V ) = max E u ( B (v + G ))
H
T

H H

H H

H*
T

1
2
N
where H = H := ( H , H ,...H ) : H is a matrix valued predictable process

H is the vector corresponding to the number of units to invest in the risky assets.
0
Having obtained H max , H can then be found by using the fact that

must be self-financing and that V0

( H 0 , H max )

= v.

We shall next consider three approaches to find an optimal portfolio.

4.5.1.

THREE SOLUTION METHODS TO FIND AN OPTIMAL

PORTFOLIO
Using differential calculus

Example 4.17

Let K1min = 1, K1max = 0 , K 2min = 1, K 2max = 0 , T = 2, N = 1 and suppose that the interest
rate which describes the evolution of the riskless asset is r. The prices of the risky
asset are given in the table below:
St1 ( )

t=0

t=1

t=2

P {}

124

1 = ( 1,0 )

10

12

14

1/4

2 = ( 1, 1)

10

12

10

1/4

3 = ( 0,0 )

10

11

1/4

4 = ( 0, 1)

10

1/4

Suppose that the utility function is given by


u ( x ) = e x

Let

10

h1 R be the number of units carried forward from time 0 to time 1

(i.e. when the price of the risky asset is 10 for all )


Let

12

h2 R be the number of units carried forward from time 1 to time 2

when the asset price is 12 (i.e. for = (-1,0) or (-1,-1))


Let 8 h2 R be the number of units carried forward from time 1 to time 2
when the asset price is 8 (i.e. for = (0,0) or (0,-1))
Then
H (1 ) = ( 10 h1 ,12 h2 )
H (2 ) = ( 10 h1 ,12 h2 )
H (3 ) = ( 10 h1 ,8 h2 )
H (4 ) = ( 10 h1 ,8 h2 )

The changes in the discounted asset prices from time zero to time one and time one
to time two are given below :

125

S11 * (1 ) =

S 21 * (1 ) =

2 10r
2 10r
2 10r
2 10r
, S11 * (2 ) =
, S11 * (3 ) =
, S11 * (4 ) =
1+ r
1+ r
1+ r
1+ r

2 12r

(1 + r )

, S 21 * (2 ) =

2 12r

(1 + r )

, S 21 * (3 ) =

3 8r

(1 + r )

, S 21 * (4 ) =

3 8r

(1 + r )

So

(1+r )2 (v+HS1* )
E u B2 (v + G2H* ) = E e

1+r 2 v+ 10 h 210 r + 12 h 212 r


1+r 2 v+ 10 h 210 r + 12 h 212 r

( )
( )



1
2
1
2

2
2

1+ r
1+r
1
1
(1+ r )

(1+ r )

= e
+
+ 4 e
4

1+r 2 v+ 10 h 210 r + 8 h 38 r
1+r 2 v+ 10 h 210 r + 8 h 38 r

( )

( )


1
2
1
2

2
2

1+r
1+ r
1
1
(1+r )
(1+ r )

+ e

e
4

1 1+r 2 v10 h1(1+r )(210 r )12 h2 (212 r ) 1 (1+r )2 v10 h1(1+r )(210 r )12 h2 (212 r )
= e ( )
e

4
2
2
1 1+r v10 h1(1+r )(210 r ) 8 h2 (38 r ) 1 (1+r ) v10 h1(1+r )(210 r ) 8 h2 (38 r )
e ( )
e

Define the function


: R3 R
1 1+r 2 vx 1+r 210 r )x2 (212 r ) 1 (1+r )2 vx1(1+r )(210 r )x2 (212 r )
( x1 , x2 , x3 ) = e ( ) 1( )(
e

4
2
2
1 1+r vx 1+r 210 r )x3 (38 r ) 1 (1+r ) vx1(1+r )(210 r )x3 (38 r )
e ( ) 1( )(
e

Then

126

0=

( 10 h 1 , 12 h 2 , 8 h 2 )
x1

(1 + r )(2 10 r )

2
(1 + r )(2 10 r ) (1+ r )2 v 10 h 1 (1+ r )(210 r ) 12 h 2 ( 212 r )
(1+ r ) v 10 h 1 (1+ r )(2 10 r ) 12 h 2 (2 12 r )
e
+
e
+

4
4
2
2
(1 + r )( 2 10 r ) (1+ r ) v 10 h 1 (1+ r )( 210 r ) 8 h 2 (38 r ) (1 + r )( 2 10 r ) (1+ r ) v 10 h 1 (1+ r )( 210 r ) 8 h 2 ( 38 r )
e
+
e

4
4

(2)
0=
=

( 10h 1 , 12h 2 , 8h 2 )
x2

2 12r (1+r )2 v10 h 1(1+r )(210 r )12 h 2 (212 r ) (2 12r ) (1+r )2 v10 h 1(1+r )(210 r )12 h 2 (212 r )
e
+
e

4
4

(3)
0=
=

( 10h 1 , 12h 2 , 8h 2 )
x3

3 8r (1+r )2 v10 h 1(1+r )(210 r ) 8 h 2 (38 r ) (3 8r ) (1+r )2 v10 h 1(1+r )(210 r ) 8 h 2 (38r )
e
+
e

4
4

(4)
From (3) and (4) we have
2

(1+ r ) v10 h 1(1+ r )( 210 r )

(2 +12r ) e

12

h 2 (2+12 r )

(1+ r ) v 10 h 1(1+r )(210 r )

=e

(2 12r ) e

12

h 2 (212 r )

and
2

(1+ r ) v10 h 1(1+r )(210 r )

(3 + 8r ) e h (3+8r ) = e(1+r ) v
2

10

h 1 (1+ r )(210 r )

(3 8r ) e h (38r )
2

respectively. Taking logs on both sides of the above and rearranging terms gives
12

1 2 12r
h 2 = ln
4 2 +12r

and
8

1 3 8r
h 2 = ln
6 3 + 8r

127

Substitute these values in (2) to obtain 10h 1 . Having obtained the number of units
to carry forward from time t to t+1 in the risky asset it remains to find the number
of units to carry forward in the riskless asset.
Let

10

h10 be the number of units carried forward from time zero to time one in the

riskless asset when the price of the risky asset is 10.


To find

10

h10 use

v = 10h10 B0 + 10h1 S01 ( )

12 0
Now let h2 be the number of units carried forward from time one to time two in
8 0
the riskless asset, if the price of the risky asset at time one is 12 and h2 be the

number of units carried forward from time one to time two in the riskless asset, if
the price of the risky asset at time one is 8.

12 0
8 0
To find h2 and h2

work out (for each ) the value of the portfolio at time one :
V1H (1 ) = 10h10 (1 + r ) + 12 10h1 = V1H (2 )
V1H (3 ) = 10h10 (1 + r ) + 8 10h1 = V1H (4 )

since H must be self financing then

V1H (1 ) = 12h20 (1 + r ) + 12 12h2 = V1H (2 )


V1H (3 ) = 8h20 (1 + r ) + 8 8h2 = V1H (4 )

128

Using Dynamic Programming

If the number of strategies and time period increases the previous approach for
finding the optimal trading strategies becomes intractable.

Another approach

which finds an optimal strategy is dynamic programming. This procedure consists


of breaking down the problem into smaller decision problems which are easier to
deal with. These sub problems correspond to single period optimization problems
and hence find an optimal control locally; that is over some subset of the original
time horizon. What optimizes at a local level, however, need not do likewise
globally, but in this case the transition from a local optimum to a global one can be
achieved through the classical Bellmans Principle of Optimality.
Bellmans Principle of Optimality states that if H* is an optimal strategy operating
from time t=0 to time t=T, then any optimal strategy operating over {s,s+1,T}
for any s=1,2,T will coincide with H* whenever the state of the system at time s
is the same for the two strategies.
Through this principle one can embed the original problem into a larger one by
varying the initial time s and initial wealth v. Hence, a family of optimization
problems parameterized by the initial pair (s,v) can be considered.
To

this

end,

for

each

let

us

introduce

the

value

function

V : {0,1,...T } :

H max *

Vs ,v () = E u BT + Gs ,T Fs () s = 0,1, 2,...T 1

Bs
VT ,v () = u (v, )

129

H*
where Gs ,T represent the discounted gains at time T if time starts from s. (Note that
H*
G0,T
= GTH* ). This function is the optimal cost-to-go over the period {s,s+1,.T}

given that the current time is s and the wealth at this time instant is v.
We claim that this function satisfies the following recursive equation:

Vs ,v () = maxN E V h Fs ()
h
s +1,V s+1

VT ,v () = u (v, )
Where

h
V s +1 = Bs +1 + hiS s*+1
Bs

with v being the wealth obtained at time s from any


h

strategy H implemented in the time period {0,1,2,s}, and V s +1 is the time s+1
wealth if the wealth at time s is v. From the above discussion if we know how to
optimally invest from time s+1 to time T, given any wealth v at time s, then at time

s we only need to find the optimal amount to invest from time s to s+1.

Proposition 4.18
Let V : {0,1,...T }

be defined using the following backward

algorithm:

VT , v () = u (v, )
Vs ,v () = maxN E V h Fs () for s = 0,1,...T 1
h
s +1,V s+1

Then for every v,

V0,v = max E u BT v + GTH*


H H

( (

))

130

By the above proposition, the multiperiod problem is reduced to solving a


sequence of single-period problems by working backwards in time.

This

procedure is known as the backward algorithm. Indeed, if s = T then by definition


VT ,v () = u (v, ) for each v ,

If s = T-1 then VT 1,v () can be found by solving


VT 1,v () = maxN E VT ,V h ()
T

(5)
We can continue in this manner until we obtain V0,v which is the value we are
interested in. For simplicity of notation we shall sometimes write V ( s, v) as the
value of Vs ,v () for a given .

Example 4.19

Let K1min = 1, K1max = 0 , K 2min = 1, K 2max = 0 , T = 2, N = 1 and suppose that the interest
rate which describes the evolution of the riskless asset is r. The prices of the risky
asset are given in the table below :
St1 ( )

t=0

t=1

t=2

P {}

1 = ( 1,0 )

10

12

14

1/4

2 = ( 1, 1)

10

12

10

1/4

3 = ( 0,0 )

10

11

1/4

4 = ( 0, 1)

10

1/4

131

The discounted changes in asset prices are:


S11 * (1 ) =

2 10r
2 10r
2 10r
2 10r
, S11 * (2 ) =
, S11 * (3 ) =
, S11 * (4 ) =
1+ r
1+ r
1+ r
1+ r

S 21 * ( 1 ) =

2 12r

(1 + r )

, S 21 * (2 ) =

2 12r

(1 + r )

, S 21 * ( 3 ) =

3 8r

(1 + r )

, S 21 * (4 ) =

3 8r

(1 + r )

Suppose that the utility function is given by u ( x) = e x . If


S11 ( ) = 12 (i.e. is 1 or 2 ) , then

S 1 S 1

2
1

V (1, v ) = max E u B2 + h { : S1 () = 12}


h
B2 B1
B1

212 r
2 v
(1+r ) 1+r +h(1+r )2


1
1

e
P { : S2 () = 14} { : S1 () = 12}

= max

(1+r )2 v +h212 r

1+r
2

(1+r )

e
P { : S21 () = 10} { : S11 () = 12}


1 1+r v h 212 r )
h 2+12 r )
= max e ( ) e (
+e (

Letting
1 :
1 1+r w x 212 r )
x 2+12 r )
+e (
1 ( x ) = e ( ) e (

we have that
1
1
(h) = e(1+r)w (2 12r ) eh(212 r ) + (2 +12r ) eh(2+12 r) = 0
x
2
1 2 12r
12
h = ln
(compare
with
h2 in the previous example)

4 2 + 12r

Suppose S11 ( ) = 8 (i.e. is 3 or 4 ) . Then

132

S 1 S 1

2
1

V (1, v ) = max E u B2 + h { : S1 () = 8}
h
B2 B1
B1

38 r
2 v
(1+r ) 1+r +h(1+r )2


1
1

e
P { : S2 () = 11} { : S1 () = 8}

= max

(1+r )2 v +h38 r

1+r
2

(1+r )

e
P { : S 21 () = 5} { : S11 () = 8}


1 1+r v h 38 r
h 3+8 r
= max e ( ) e ( ) + e ( )

Let
2 :
1 1+r v x 38 r
x 3+8 r
2 ( x ) = e ( ) e ( ) + e ( )

Then we get that


2
1
(h) = e(1+r)v (3 8r ) eh(38r) + (3 + 8r ) eh(3+8r ) = 0
2
x
1 3 8r
8
h = ln
(compare
with
h2 in the previous example)

6 3 + 8r

So
212 r 212 r
2+12 r 212 r
ln

1 (1+r )v 4 ln 2+12 r
2+12 r
4
V1,v () = e
e
+
e

where =1 , 2
2

38 r 38 r
3+8 r 38 r
ln

1 (1+r )v 6 ln 3+8r
3+8 r
6
V1,v = e
e
+
e

where = 3 , 4
2

( )

Now let t = 0 so that S01 ( ) = 10 for all . By the dynamic programming equation
we obtain:

133

V (0, v) = max E V 1,V1h { : S01 () = 10}


h

v
S 1 S 1

max E V 1, B1 + h 1 0 { : S01 () = 10}


h
B1 B0
B0

v
S1 S1

1
0

r
2+12 r 212 r
ln

1 (1+r )B1 B0 +h B1 B0 2412 r ln 22+12


2+12 r
12 r
4
e
+
e
e

= max

1
h
3+8 r 38 r
1 (1+r )B1 Bv +h SB1 BS0 38 r ln 38 r

ln

1
0
0
3+8 r
3+8 r
6
6
e

e
+
e
4


v
v
S1 S1
S1 S1

1
0
0

1
1 (1+r )B1 B0 +h B1 B0
1 (1+r )B1 B0 +h B1 B0
= max e
f ( ) e
f
h
4
4

( )

( )

where
2+12 r 212 r
212 r ln 212 r
e 4 2+12 r + e 4 ln 2+12 r if =
f () =
38 r 38r
3+8 r 38 r
ln
ln

3+8 r
+
r
6
3
8
6
e
+
e
if =

Letting

3 :
2
2
1
1
3 ( x) = e(1+r ) v(1+r )(210 r ) x f () e(1+r ) v(1+r )(210 r ) x f ()
4
4

We have that
2
3
1
(h) = (1 + r )(2 10r ) f () e(1+r) w(1+r)(210 r)h +
4
x
2
^
1
(1 + r )(2 10r ) f e(1+r) w(1+r)(210 r)h = 0

4

f ()
1 2 10r
h=
ln
+ ln ^
20r 2 + 10r
f

134

Note that h should be equivalent to 10h1 in the previous example.


To find the number of units to invest in the riskless asset use the same procedure as
in the previous example.

Using the Martingale Method

Another method which generalizes easily from the single period to the multiperiod
case is the martingale approach. The only difference between the single period and
the multiperiod case is the discounting term. Instead of B1 we have to use BT.
Otherwise the equations involved remain the same.
So Wv is now defined as follows:

Wv = {random variables W : R such that E Q [W / BT ] = v}


Thus any contingent claim W is replicable by corresponding portfolio H, with

VTH ( ) = W ( ) , and by the risk neutral valuation principle, for any trading
strategy H with initial value v, E Q [W / BT ] = v
Again using Lagrangian multipliers, maximizing

E[u (W )] subject to the

constraint W Wv becomes :
Max E[u (W )] E Q [W / BT ] v

Q ( )
L
(

)
=
Let
P ( ) and ignoring v (being a constant) we have

135

E[u (W )] E[ LW / BT ] = E[u (W ) LW / BT ] =

{u(W ()) L()W () / B }P()


T

u
The necessary differentiability conditions give x (W ( )) L( ) / BT = 0 . But
u
again with x and L always positive,

u
(W ( )) = L( ) / BT for each .
x

Otherwise the derivative cannot be zero at any .


These conditions can be rearranged as:

u
(W ( )) = L( ) / BT = Q( ) /( BT P( ))
x
which give

Q( ) =

u
(W ( )) BT P( ) /
x

But under optimality, (as shown in chapter two with BT replacing B1)

u ( BT VT ( ), )
P( )

x
Q( ) =
u ( BT VT )
E

forcing = E BT u (W )

x .

Denoting by I the inverse to u with respect to its first variable, we have :

u (W ( ), ))
I
= W ( )

136

so

L( ))
u(W ( ), ))

W ( ) = I
= I

x
BT

To obtain an explicit formula for we see that for the contingent claim W,
E Q [W / BT ] = v

which in our case translates to the equation:

EQ I
BT = v
BT

This equation has a unique solution because of the nature of the function I. The
expected value of W will give the optimal wealth. Being in a complete market we
can work out the corresponding portfolio H.

Example 4.20

Let K1min = 1, K1max = 0 , K 2min = 1, K 2max = 0 , T = 2, N = 1 and suppose that the interest
rate which describes the evolution of the riskless asset is r. The prices of the risky
asset are given in the table below :
St1 ( )

t=0

t=1

t=2

P {}

1 = ( 1,0 )

10

12

14

1/4

2 = ( 1, 1)

10

12

10

1/4

3 = ( 0,0 )

10

11

1/4

137

4 = ( 0, 1)

10

1/4

Suppose that the utility function is given by


u ( x ) = e x

Compute Q , and I
o Q was already obtained in the previous chapter by using the
*
*
martingale property E Q [ St | Fs ] = S s , s < t :

1 + 5r 1 + 6 r
Q(1 ) =

2 2
1 + 5r 1 6 r
Q(2 ) =

2 2
1 5r 1 + 4 r
Q(3 ) =

2 2
1 5r 1 4 r
Q(4 ) =

2 2
Q( )
(
)
L

=
o
P ( )

L (1 ) = (1 + 5r )(1 + 6r ) , L(2 ) = (1 + 5r )(1 6r )


L (3 ) = (1 5r )(1 + 4r ) , L (4 ) = (1 5r )(1 4r )

( )

o I ( y ) = ln y

138

E
I
B

Obtain by using Q B 1 = v
1

=e

ln B2 EQ [ln L ] B2 v

Compute the optimal wealth W:


L())
= B2v + E Q [ln L ] ln L
W ( ) = I
B2

ln (1 + 5r )(1 + 6r ) + ln (1 + 5r )(1 6r ) + ln (1 5r )(1 + 4r ) + ln (1 5r )(1 4r )


W (1 ) = B2v +
ln (1 + 5r )(1 + 6r )
4
ln (1 + 5r )(1 + 6r ) + ln (1 + 5r )(1 6r ) + ln (1 5r )(1 + 4r ) + ln (1 5r )(1 4r )

W (2 ) = B2v +
ln (1 + 5r )(1 6r )
4

ln (1 + 5r )(1 + 6r ) + ln (1 + 5r )(1 6r ) + ln (1 5r )(1 + 4r ) + ln (1 5r )(1 4r )


ln (1 5r )(1 + 4r )
W (3 ) = B2v +
4
ln (1 + 5r )(1 + 6r ) + ln (1 + 5r )(1 6r ) + ln (1 5r )(1 + 4r ) + ln (1 5r )(1 4r )

ln 1 5r 1 4r
W (4 ) = B2v +
(
)(
)
4

Find the portfolio H which replicates W


Let 10h10, 10h1, 12h20 , 8h20, 12h2 , 8h2 be defined as in the example in section 4.1.1.1.
12 0 8 0 12
8
2
2
2
2

h , h , h, h

can be found by solving the following system of linear

equations
2 12 0
2

h + 14 12h2

W (1 ) = (1 + r )

2 12 0
2

W (2 ) = (1 + r )

h + 10 12h2

W (3 ) = (1 + r ) 8h20 + 118h2
2

W (4 ) = (1 + r ) 8h20 + 15 8h2
10 0
1

h and 10h1 can then be found by solving the following system of linear

equations (recalling the self-financing property of H)


(1 + r ) 10h10 + 12 10h1 = V1 (1 ) = (1 + r ) 12h20 + 12 12h2 (or 2 )
(1 + r ) 10h10 + 8 10h1 = V1 (3 ) = (1 + r ) 8h20 + 8 8h2 (or 4 )

139

Das könnte Ihnen auch gefallen