This action might not be possible to undo. Are you sure you want to continue?

Welcome to Scribd! Start your free trial and access books, documents and more.

Find out moreUndergraduate Introduction to

Mathematics

J Robert Buchanan

An

Undergraduate to

{f^ I n t r o d u c t i o n

Financial^! Mathematics

This page is intentionally left blank

•M'!'M>»M4

J Robert Buchanan

MiNersviile University, USA

World Scientific

NEW JERSEY • LONDON - SINGAPORE • BEIJING • SHANGHAI • HONG KONG • TAIPEI • CHENNAI

Published by World Scientific Publishing Co. Pte. Ltd. 5 Toh Tuck Link, Singapore 596224 USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601 UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE

British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library.

AN UNDERGRADUATE INTRODUCTION TO FINANCIAL MATHEMATICS Copyright © 2006 by World Scientific Publishing Co. Pte. Ltd. All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means, electronic or mechanical, including photocopying, recording or any information storage and retrieval system now known or to be invented, without written permission from the Publisher.

For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy is not required from the publisher.

ISBN 981-256-637-6

Printed in Singapore by World Scientific Printers (S) Pte Ltd

Dedication For my wife. . Monika.

This page is intentionally left blank .

One of my objectives in writing this book was to create a readable. I stand in awe of authors such as those. The book assumes no background on the part of the reader in probability or statistics.Preface This book is intended for an audience with an undergraduate level of exposure to calculus through elementary multivariable calculus. I have taught such a course now three times and this book grew out of my lecture notes and reading for the course. though the well-prepared reader may be able to skip the first several with no loss of understanding in what comes later. The second and third chapters provide an introduction to the concepts of probability and statistics which will be used throughout the remainder of the book. This book consists of ten chapters which are intended to be read in order. Chapter Three introduces continuous random variables and emphasizes the similarities and differences between discrete and continuous random variables. Both discretely compounded and continuously compounded interest are treated there. The book begins with the theory of interest because this topic is unlikely to scare off any reader no matter how long it has been since they have done any formal mathematics. Knowing the amount of work required to produce this book. The nor- vii . My desire to write such a book grew out of the need to find an accessible book for undergraduate mathematics majors on the topic of financial mathematics. Chapter Two deals with discrete random variables and emphasizes the use of the binomial random variable. reasonably self-contained introduction to financial mathematics for people wanting to learn some of the basics of option pricing and hedging. The first chapter is on interest and its role in finance. so in the time it took me to compose this book there may have appeared several superior works on the subject. New titles in financial mathematics appear constantly.

For readers already well versed in calculus. The method I choose to solve the PDE involves the use of the Fourier Transform. I have attempted to introduce stochastic processes in an intuitive manner and by connecting elementary stochastic models of some processes to their corresponding deterministic counterparts. Properties of options such as the Put/Call Parity formula are presented and justified. this is the first material which may be unfamiliar to them. Both European and American style options are discussed though the emphasis is on European options. It also provides exposure to the Fourier . The assumption that financial calculations are carried out in an "arbitrage free" setting pervades the remainder of the book. The lack of arbitrage opportunities in financial transactions ensures that it is not possible to make a risk free profit. This derivation makes use of the earlier material on arbitrage. Most three. Ito's Lemma is introduced and an elementary proof of this result is given based on the multivariable form of Taylor's Theorem. In the fourth chapter the concept of arbitrage is introduced. Thus this chapter begins with a brief discussion of the Fourier and Inverse Fourier Transforms and their properties.viii An Undergraduate Introduction to Financial Mathematics mal random variable and the close related lognormal random variable are introduced and explored in the latter chapter. The choice of material to present and the method of presentation is difficult in this chapter due to the complexities and subtleties of stochastic processes. Chapter Six introduces the topic of options. thus students will have the calculus background necessary to follow this discussion. Readers whose interest is piqued by material in Chapter Five should consult the bibliography for references to more comprehensive and detailed discussions of stochastic calculus. stochastic processes and the Put/Call Parity formula. There are several different methods commonly used to derive the solution to the PDE and students benefit from different aspects of each derivation. The fifth chapter introduces the reader to the concepts of random walks and Brownian motion. and statistics. The random walk underlies the mathematical model of the value of securities such as stocks and other financial instruments whose values are derived from securities. The seventh chapter develops the solution to the Black-Scholes PDE.or foursemester elementary calculus courses include at least an optional section on the Fourier Transform. This chapter includes a discussion of the result from linear algebra and operations research known as the Duality Theorem of Linear Programming. probability. In this chapter we also derive the partial differential equation and boundary conditions used to price European call and put options.

and the volatility of the underlying security's value. Some of the classical models of portfolio selection are introduced in this chapter including the Capital Assets Pricing Model (CAPM) and the Minimum Variance Portfolio. Mark Elicker. Stephen Kluth. The tenth chapter discusses several different notions of optimality in selecting portfolios of investments. Chapter Eight introduces some of the commonly discussed partial derivatives of the Black-Scholes option pricing formula. Alicia Kasif. and the volatility of the underlying security's value. Joshua Wise. A list of errata and other information related to this book can be found at a web site I created: .Preface IX Transform for students who will be later taking a course in PDEs and more importantly exposure for students who will not take such a course. The Greeks are used in the ninth chapter on hedging strategies for portfolios. Chapter Nine will discuss and illustrate several examples of hedging strategies. Jessica Paxton. The author is indebted to the students of that class for finding numerous typographical errors in that earlier version which were corrected before the camera ready copy was sent to the publisher. Pamela Wentz. After completing this derivation of the Black-Scholes option pricing formula students should also seek out other derivations in the literature for the purposes of comparison. Christopher Rachor. Timothy Ren. Jennifer Gomulka. the risk-free interest rate. and Michael Zrncic. The collection of partial derivatives introduced in this chapter is commonly referred to as "the Greeks" by many financial practitioners. Patrick McDevitt. These partial derivatives help the reader to understand the sensitivity of option prices to movements in the underlying security's value. the risk-free interest rate. Jason Buck. Nicole Hundley. Mathematically the hedging strategies remove some of the low order terms from the Black-Scholes option pricing formula making it less sensitive to changes in the variables upon which it depends. Hedging strategies are used to protect the value of a portfolio against movements in the underlying security's value. but intellectually rewarding work. Chapter Ten extends the ideas introduced in Chapter Nine by modeling the effects of correlated movements in the values of investments. It is the author's hope that students will find this book a useful introduction to financial mathematics and a springboard to further study in this area. Kelly Flynn. During the summer of 2005 a draft version of this manuscript was used by the author to teach a course in financial mathematics. The author wishes to thank Jill Bachstadt. Writing this book has been hard.

USA October 31. PA.edu/~bob/book/ Please feel free to share your comments. J.millersville. and (I hope) praise for this work through the email address that can be found at that site. criticism. Robert Buchanan Lancaster.x An Undergraduate Introduction to Financial Mathematics http://banach. 2005 .

8 Events and Probabilities Addition Rule Conditional Probability and Multiplication Rule Random Variables and Probability Distributions Binomial Random Variables Expected Value Variance and Standard Deviation Exercises 3.2 2.3 2.1 1.1 2.2 1.6 2.2 3.4 3.4 2.5 Continuous Random Variables Expected Value of Continuous Random Variables Variance and Standard Deviation Normal Random Variables Central Limit Theorem xi .5 1.4 1.3 1. Simple Interest Compound Interest Continuously Compounded Interest Present Value Rate of Return Exercises vii 1 1 3 4 5 11 12 15 15 17 18 21 23 24 29 32 35 35 38 40 42 49 Discrete Probability 2.Contents Preface 1.6 2.1 3.5 2.3 3. The Theory of Interest 1. Normal Random Variables and Probability 3.7 2.

Solution of the Black-Scholes Equation 7.3 6.2 7.5 Properties of Options Pricing an Option Using a Binary Model Black-Scholes Partial Differential Equation Boundary and Initial Conditions Exercises 7. Options 6.7 Intuitive Idea of a Random Walk First Step Analysis Intuitive Idea of a Stochastic Process Stock Market Example More About Stochastic Processes Ito's Lemma Exercises 6. Lognormal Random Variables Properties of Expected Value Properties of Variance Exercises 51 55 58 61 63 63 64 66 72 74 77 77 78 91 95 97 98 101 103 104 107 110 112 114 115 115 118 119 122 127 131 131 133 The Arbitrage Theorem 4.2 4. Derivatives of Black-Scholes Option Prices 8.2.2 5.8 3.4 5.3 7.4 7.4 The Concept of Arbitrage Duality Theorem of Linear Programming 4.5 Fourier Transforms Inverse Fourier Transforms Changing Variables in the Black-Scholes PDE Solving the Black-Scholes Equation Exercises 8.6 3.9 4.1 4.2 6.1 Dual Problems The Fundamental Theorem of Finance Exercises 5.3 5.1 7.5 5.7 3.4 6.1 5. Random Walks and Brownian Motion 5.1 6.3 4.2 Theta Delta .6 5.Xll An Undergraduate Introduction to Financial Mathematics 3.1 8.

Contents

xiii

8.3 8.4 8.5 8.6 8.7 9.

Gamma Vega Rho Relationships Between A, 9 , and T Exercises

135 136 138 139 141 143 143 145 149 151 153 155 155 164 165 171 173 177 186 191 195 203 203 206 212 225 231 235 239 245 249 255 265 267

Hedging 9.1 9.2 9.3 9.4 9.5 General Principles Delta Hedging Delta Neutral Portfolios Gamma Neutral Portfolios Exercises

10. Optimizing Portfolios 10.1 10.2 10.3 10.4 10.5 10.6 10.7 10.8 Covariance and Correlation Optimal Portfolios Utility Functions Expected Utility Portfolio Selection Minimum Variance Analysis Mean Variance Analysis Exercises Sample Stock Market Data Solutions to Chapter Exercises

Appendix A Appendix B B.l B.2 B.3 B.4 B.5 B.6 B.7 B.8 B.9 B.10

The Theory of Interest Discrete Probability Normal Random Variables and Probability The Arbitrage Theorem Random Walks and Brownian Motion Options Solution of the Black-Scholes Equation Derivatives of Black-Scholes Option Prices Hedging Optimizing Portfolios

Bibliography Index

Chapter 1

The Theory of Interest

One of the first types of investments that people learn about is some variation on the savings account. In exchange for the temporary use of an investor's money, a bank or other financial institution agrees to pay interest, a percentage of the amount invested, to the investor. There are many different schemes for paying interest. In this chapter we will describe some of the most common types of interest and contrast their differences. Along the way the reader will have the opportunity to renew their acquaintanceship with exponential functions and the geometric series. Since an amount of capital can be invested and earn interest and thus numerically increase in value in the future, the concept of present value will be introduced. Present value provides a way of comparing values of investments made at different times in the past, present, and future. As an application of present value, several examples of saving for retirement and calculation of mortgages will be presented. Sometimes investments pay the investor varying amounts of money which change over time. The concept of rate of return can be used to convert these payments in effective interest rates, making comparison of investments easier.

1.1

Simple Interest

In exchange for the use of a depositor's money, banks pay a fraction of the account balance back to the depositor. This fractional payment is known as interest. The money a bank uses to pay interest is generated by investments and loans that the bank makes with the depositor's money. Interest is paid in many cases at specified times of the year, but nearly always the fraction of the deposited amount used to calculate the interest is called the interest rate and is expressed as a percentage paid per year.

1

2

An Undergraduate Introduction

to Financial

Mathematics

For example a credit union may pay 6% annually on savings accounts. This means that if a savings account contains $100 now, then exactly one year from now the bank will pay the depositor $6 (which is 6% of $100) provided the depositor maintains an account balance of $100 for the entire year. In this chapter and those that follow, interest rates will be denoted symbolically by r. To simplify the formulas and mathematical calculations, when r is used it will be converted to decimal form even though it may still be referred to as a percentage. The 6% annual interest rate mentioned above would be treated mathematically as r — 0.06 per year. The initially deposited amount which earns the interest will be called the principal amount and will be denoted P. The sum of the principal amount and any earned interest will be called the compound amount and A will represent it symbolically. Therefore the relationship between P, r, and A for a single year period is A = P + Pr = P(l + r). The interest, once paid to the depositor, is the depositor's to keep. Banks and other financial institutions "pay" the depositor by adding the interest to the depositor's account. Unless the depositor withdraws the interest or some part of the principal, the process begins again for another interest period. Thus two interest periods (think of them as years) after the initial deposit the compound amount would be A = P(l + r) + P ( l + r)r = P{1 + r)2. Continuing in this way we can see that t years after the initial deposit of an amount P , the compound amount A will grow to A = P{l + r)t (1.1)

This is known as the simple interest formula. A mathematical "purist" may wish to establish Eq. (1.1) using the principle of induction. Banks and other interest-paying financial institutions often pay interest more than a single time per year. The simple interest formula must be modified to track the compound amount for interest periods of other than one year.

The Theory of Interest

3

1.2

Compound Interest

The typical interest bearing savings or checking account will be described an investor as earning a specific annual interest rate compounded monthly. In this section will be compare and contrast compound interest to the simple interest case of the previous section. Whenever interest is allowed to earn interest itself, an investment is said to earn compound interest. In this situation, part of the interest is paid to the depositor more than once per year. Once paid, the interest begins earning interest. We will let the number of compounding periods per year be n. For example for interest "compounded monthly" n = 12. Only two small modifications to the simple interest formula (1.1) are needed to calculate the compound interest. First, it is now necessary to think of the interest rate per compounding period. If the annual interest rate is r, then the interest rate per compounding period is r/n. Second, the elapsed time should be thought of as some number of compounding periods rather than years. Thus with n compounding periods per year, the number of compounding periods in t years is nt. Therefore the formula for compound interest is

A = P(l+r-)a'.

(1.2)

Example 1.1 Suppose an account earns 5.75% annually compounded monthly. If the principal amount is $3104 then after three and one half years the compound amount will be 0 0575\ ( 1 2 ) ( 3 - 5 ) A = 3104(1 + H ^ I M =3794.15.

The reader should verify that if the principal in the previous example earned a simple interest rate of 5.75% then the compound amount after 3.5 years would be only $3774.88. Thus happily for the depositor, compound interest builds wealth faster than simple interest. Frequently it is useful to compare an annual interest rate with compounding to an equivalent simple interest, i.e. to the simple annual interest rate which would generate the sample amount of interest as the annual compound rate. This equivalent interest rate is called the effective interest rate. For the amounts and rates mentioned in the previous example we can find the effective interest

4

An

Undergraduate

Introduction

to Financial

Mathematics

rate by solving the equation

3104 (l+°-^y

= 3104(1 + re)

1.05904 = 1 + re 0.05904 = re

Thus the annual interest rate of 5.75% compounded monthly is equivalent to an effective annual simple rate of 5.904%. Intuitively it seems that more compounding periods per year implies a higher effective annual interest rate. In the next section we will explore the limiting case of frequent compounding going beyond semiannually, quarterly, monthly, weekly, daily, hourly, etc. to continuously.

1.3

Continuously Compounded Interest

Mathematically when considering the effect on the compound amount of more frequent compounding, we are contemplating a limiting process. In symbolic form we would like to find the compound amount A which satisfies the equation

/

r

\nt

A=

lim P 1 + n—>oo V n/

•

(1.3)

Fortunately there is a simple expression for the value of the limit on the right-hand side of Eq. (1.3). We will find it by working on the limit

lim f l + - ) " .

n—too V 7J/

This limit is indeterminate of the form 1°°. We will evaluate it through a standard approach using the natural logarithm and l'HopitaPs Rule. The reader should consult an elementary calculus book such as [Smith and Minton (2002)] for more details. We see that if y = (1 + r/n)n, then lny = m ( l + - ) = n l n ( l + r/n) _ ln(l + r/n) ~ ljn which is indeterminate of the form 0/0 as n — oo. To apply l'Hopital's Rule > we take the limit of the derivative of the numerator over the derivative of

34305%. Later in this book present and future values will help us determine a fair price for stock market derivatives. A = Pert. This is the comparison between the present value of an investment versus its future value.15% compounded continuously. Finally we arrive at the formula for continuously compounded interest. We will see in this section that present and future value play central roles in planning for retirement and determining loan payments.5) = 4180. Thus lim In y = lim £(ln(l + r/n)) r = lim . Example 1.4) This formula may seem familiar since it is often presented as the exponential growth formula in elementary algebra.2 Suppose $3585 is deposited in an account which pays interest at an annual rate of 6. (1. or calculus.The Theory of Interest 5 the denominator. precalculus.4 Present Value One of the themes we will see many times in the study of financial mathematics is the comparison of the value of a particular investment at the present time with the value of the investment at some point in the future. n^oo 1 + r/n = r £ (Vn) Thus limn^oo y — er.0615 = 1 + r e which implies re = 6. . The effective simple interest rate is the solution to the equation e0. After two and one half years the principal plus earned interest will have grown to A = 3585e (0 ' 0615)(2 .82. . 1. The quantity A has the property that A changes with time t at a rate proportional to A itself.

In other words if an investor wishes to have A monetary units in savings t years from now and they can place money in a savings account earning interest at an annual rate r compounded continuously.99% compounded monthly. (1.3 Suppose an investor will receive payments at the end of the next six years in the amounts shown in the table. the future value of P is just the compound amount of P monetary units invested in a savings account earning interest r compounded continuously for t years. There are also formulas for future and present value when interest is compounded at discrete intervals. Year Payment 1 2 3 4 5 6 465 233 632 365 334 248 If the interest rate is 3.4). what is the present value of the investments? Assuming the first payment will arrive one year from . If the interest rate is r annually with n compounding periods per year then the future value of P is Compare this equation with Eq. Thus by comparison with Eq. the case of simple interest can be handled.2). By contrast the present value of A in an environment of interest rate r compounded continuously for t years is P = Ae~rt. Example 1. Simple algebra shows then the present value of P earning interest at rate r compounded n times per year for t years is By choosing n = 1. the investor should deposit P monetary units now.6 An Undergraduate Introduction to Financial Mathematics The future value t years from now of an invested amount P subject to an annual interest rate r compounded continuously is A = Pert. not continuously. (1.

an+1 S ~ 1-a provided a ^ l . Loans are always made under the assumptions of a prevailing interest rate (with compounding). Now we will apply this tool to the task of finding out the monthly amount of a loan payment. Usually portions of the loan must be repaid at regular intervals (for example. (1. i.(a + a2 + a3 + • • • + an+1) 5(1 .5) are multiplied by a and then subtracted from Eq. The bank issuing the automobile loan charges interest at the annual rate of r compounded n times per year.The Theory of Interest 7 now. the length of the loan. +233 + 233 { ^r) fl + ^ V 1+ + 2 4 + 632 +M2 •l+ (l + [ ^2 7 2 365(l + ^ ) " \ 3 3 4 ( l ^)"° 248(l + M f ) " = 2003. and the interest rate to calculate the loan payment. Now we turn our attention to the question of using the amount borrowed. you may some day apply for a loan from a bank or other financial institution.+ an (1. and the lifespan of the loan. Suppose someone borrows P to purchase a new car.an+1 1 . (1. Suppose the payment amount is the constant x. an amount to be borrowed. the present value is the sum 465 [ 1 + ^ V + 1 2 12 . Unless the reader is among the very fortunate few who can always pay cash for all purchases. monthly).01 Notice that the present value of the payments from the investment is different from the sum of the payments themselves (which is 2277). The length of the loan will be t years. A very helpful mathematical tool for answering questions regarding present and future values is the geometric series.o) = 1 .aS = 1 + a + a2 + • • • + an .e. If both sides of Eq.5) where n is a positive whole number. If the first . Suppose we wish to find the sum S=l + a + a2 + --. The monthly payment can be calculated if we apply the principle that the present value of all the payments made must equal the amount borrowed. the time the borrower has to repay the loan.5) we have S .

the compounding frequency. the length of the loan.8 An Undergraduate Introduction to Financial Mathematics payment must be made at the end of the first compounding period. The monthly deposit amount will be be denoted by the symbol x.6) Example 1. i . then the present value of all the payments is n x(l + . The amount of money the person should invest monthly while working can be determined by equating the present value of all their deposits with the present value of all their withdrawals. For the next 40 years they plan to invest a portion of their monthly income in securities which earn interest at the rate of 10% compounded monthly.99% compounded monthly and makes equal monthly payments. The first deposit will be made one month from now and the first withdrawal will be made 481 months from now. Similar reasoning can be used when determining how much to save for retirement.0499/12)]~ (12)(5) ) = 471. and the payment amount is expressed in the following equation.( i + ^ r « 117. the payment amount will be x = 25000(0.765a.) .0499/12) (l . Suppose a person is 25 years of age now and plans to retire at age 65.67.)\ n r + r 2 2 + •• 1 • + x(l + . The present value of all the deposits made into the retirement account is fVii°-i(v hS v) xfii^y11-^^"480 y 12. . After retirement the person plans on receiving a monthly payment (an annuity) in the absolute amount of $1500. the principal amount borrowed.-nt n = x{i = x -r l 1 - 1 -..[1 + (0.4 If a person borrows $25000 for five years at an interest rate of 4.( 1 + S ) nt i-a + -)" —n ' Therefore the relationship between the interest rate. n x— r 1+"n (1.

For the purpose of this example we will assume that the interest rate is r = 0. Waiting ten years to begin saving for retirement nearly triples the amount which the future retiree must set aside for retirement.5 Suppose two people will retire in twenty years. but such is the power of compound interest and starting a savings plan for retirement early.^ 360 1500 Y^ ( l + " j y ) i=361 ^ ' 0 10 \ ~ 8616. + ^)•— 3 6 0 . The amount they put aside during those first four years remains invested. 12 J v w 3182. monthly. 12 J l . but all other factors remain the same then X i=l 720 K. ^ ^ ^ ^ V 12 .The Theory of Interest 9 Meanwhile the present value of all the annuity payments is 840 / 1500 Y^ .03 dollars per month.94. 1+ 0. The initial amounts invested are of course invested for a longer period of time and thus contribute a proportionately greater amount to the future value of the retirement account.ioy 4 8 1 i-(i = 1500 1 + .^)-. Let us explore the difference in the final amount of retirement savings that each person will possess.61 monthly. If the person waits ten years to begin saving for retirement.36 which implies the person must invest x ~ 75. One begins saving immediately for retirement but due to unforeseen circumstances must abandon their savings plan after four years. but no additional amounts are invested during the last sixteen years of their working life.fi + ^ior Thus x w 27. The first worker has upon retirement an account whose present value is zfVl + ^5. Example 1.05 compounded monthly and that both workers will invest the same amount x. They save for retirement only during the last sixteen years of their working life. The other person waits four years before putting any money into a retirement savings account.10V* A o. This seems like a small amount to invest.

. Then once the worker is retired the monthly annuity must be discounted by the rate rs — r^.10 An Undergraduate Introduction to Financial Mathematics The present value of the second worker's total investment is ^ / l ^ [ l 0.03 for the entire lifespan of the worker/retiree.784a. The rate of inflation (usually expressed as an annual percentage rate. The first worker saves approximately 40% of what the second worker saves in only one fifth of the time..423/108. similar to an interest rate) varies with time and is a function of many factors including political.r . If the interest rate on savings is rs and the inflation rate is r^. and the rate of inflation will be ri = 0. x + Thus the second worker retires with a larger amount of retirement savings.05 V * ~ w ) ~ 108. i3 i g 48 . the ratio of their retirement balances is only 43. This can happen when wages are arbitrarily increased without an equal increase in worker productivity. Assuming the worker will make the first deposit in one month the present value of all deposits to be made is vfi i °. The economic concept of inflation is the phenomenon of the decrease in the purchasing power of a unit of money relative to a unit amount of goods or services. and international factors. Thus relative to the supply of goods. the worker will save for 40 years and live on a monthly annuity whose inflation adjusted value will be $1500 for 30 years. economic. inflation is generally described as a condition which results from an increase in the amount of money in circulation without a commensurate increase in the amount of available goods. the monthly amount put into savings by the worker must be discounted by an effective rate rs + r^. We now focus on the effect that inflation may have on the worker planning to save for retirement.fn o-i3v i-(i+ #)" ° w 91. however.102 « 0. the value of the currency is decreased. While the causes of inflation can be many and complex.40. Returning to the earlier example consider the case in which rs = 0.102a.10. The discussion of retirement savings makes no provision for rising prices.

( 1 + T ^ " 3 6 ° v = 1500 1 ' > — 12 v 7 x I1 ^ 12 J « 13822. This is not much money to live on for an entire month.07 V 4 8 1 1 .30. 1.5 Rate of Return The present value of an item is one way to determine the absolute worth of the item and to compare its worth to that of other items.07\"1 7 . the rate of return can be thought of as the interest rate per time unit that the invested amount would have to earn so that the present value of the payoff amount is equal to the invested amount.. Another way to judge the value of an item which an investor may own or consider purchasing is known as the rate of return. However. then the rate of return is r = 0..10 or 10%. / 0.. since inflation does tend to take place over the long run. In this case the rate of return per period is the interest rate such that the present value of the sequence of payoffs is equal to the amount invested.6 If you loan a friend $100 today with the understanding that they will pay you back $110 in one year's time. An} at regular intervals. In .17.. This is roughly six times the monthly investment amount when inflation is ignored. ignoring a 3% inflation rate over the lifetime of the individual would mean that the present value purchasing power of the last annuity payment would be 0.60. If a person invests an amount P now and receives an amount A one time unit from now. In a more general setting a person may invest an amount P now and receive a sequence of positive payoffs {Ai. it will be denoted by the symbol r. Then by definition P = A(l + r ) _ 1 or equivalently r = — — 1. Example 1. Thus the monthly deposit amount is $150.03^ 840 1500(1 + ——I « 184.The Theory of Interest 11 The present value of all the annuity payments is given by ^ f 1500 > i=481 v 1+ 0. Since the rate of return is going to be thought of as as equivalent interest rate.A2.

In the limit as r approaches —1 from the right. oo). Rates of return can be either positive or negative. What is the compound amount after five years? (2) Suppose that 3993 is deposited in an account which earns 4. What is the compound amount after two years if the interest is compounded .. The reader is encouraged to show that r* is unique in the exercises. 108 of [Smith and Minton (2002)]) there exists r* with .6 Exercises (1) Suppose that $3659 is deposited in a savings account which earns 6. i.2 of [Smith and Minton (2002)]) can be used to approximate the solution r* K.e. the function values approach — P < 0 asymptotically. Defining the function f(r) to be n f(r) = -P + J2Ml + r)-* i=l (1.1 < r* < oo such that f(r*) = 0. The rate of return per year is the solution to the equation. If the sum of the payoffs is less than the amount invested then /(0) < 0 and the rate of return in negative. " 10 ° + 21 TT7 + 22 (1 + r) 2 + 23 (1 + r ) 3 + 24 (1 + r ) 4 + 25 (1 + r ) 5 " ' Newton's Method (Sec. Example 1. 3. the sum of the payoffs is greater than the amount invested then r* > 0 since f(r) changes sign on the interval [0. the function values approach positive infinity.047.0].7) we can see that f(r) is continuous on the open interval (—1. 0.5% simple interest. 25}. If /(0) > 0. 22. 23.12 An Undergraduate Introduction to Financial Mathematics this case n i=i It is not clear from this definition that r has a unique value for all choices of P and payoff sequences. On the other hand as r approaches positive infinity. Thus by the Intermediate Value Theorem (p.7 Suppose you loan a friend $100 with the agreement that they will pay you at the end of the next five years amounts {21. oo). 24. In this case the function f(r) changes sign on the interval (—1. 1.3% interest.

The Theory of Interest 13 (a) (b) (c) (d) monthly? weekly? daily? continuously? (3) Find the effective annual simple interest rate which is equivalent to 8% interest compounded quarterly. If you . Beginning with Eq.3) let h = r/n. (1. Year Investment A Investment B 1 200 198 2 211 205 3 198 211 4 205 200 (8) Suppose you wish to buy a house costing $200000. (1999)]). (7) Which of the two investments described below is preferable? Assume the first payment will take place exactly one year from now and further payments are spaced one year apart. How long would it take for the compound amounts to differ by $1? (6) Many textbooks determine the formula for continuously compounded interest through an argument which avoids the use of l'Hopital's Rule (for example [Goldstein et al. while the other pays interest at an annual rate of 4.75% compounded daily. You will put a down payment of 20% of the purchase price and borrow the rest from a bank for 30 years at a fixed interest rate r compounded monthly.) n ' = P(l +/i)(1/h)rt and we can focus on finding the l i m ^ o ( l + h)1'11. Then P ( l + .75%. In order to be competitive you must meet or exceed the interest paid by another bank which pays 5.75% compounded continuously. What is the minimum interest rate you can pay and remain competitive? (5) Suppose you have $1000 to deposit in one of two types of savings accounts. (4) You are preparing to open a bank which will accept deposits into savings accounts and which will pay interest compounded monthly. Assume the continually compounded annual interest rate is 2. Show that (l + /j)l/k = e(l/h)ln(l + h) and take the limit of both sides as h — 0.25% compounded daily. Hint: you can use the > definition of the derivative in the exponent on the right-hand side. One account pays interest at an annual rate of 4.

(1.4000. which investment would you choose? Assume the simple annual interest rate is 4.7) is unique.3000. what is the maximum annual interest rate for the mortgage loan? (9) Use the Mean Value Theorem (p. (10) Suppose for an investment of $10000 you will receive payments at the end of each of the next four years in the amounts {2000. which investment would you choose? . 235 of [Stewart (1999)]) to show the rate of return defined by the root of the function in Eq. Each investment will pay you an amount listed in the table below at the end of the next five years. (b) Using the rate of return per year of the investment to make the decision.3000}. Year Investment A Investment B 1 225 220 2 215 225 3 250 250 4 225 250 5 205 210 (a) Using the present value of the investment to make the decision.14 An Undergraduate Introduction to Financial Mathematics wish your monthly mortgage payment to be $1500 or less.33%. What is the rate of return per year? (11) Suppose you have the choice of investing $1000 in just one of two ways.

In this book a probabilistic or stochastic model of a market will be developed instead. To the statistician. a numerical quantity whose value is not known until an experiment is conducted. 2.1 Events and Probabilities To the layman an event is something that happens. expected value (or mean) and standard deviation. There are many different kinds of discrete random variables. This brings up the question of what is an experiment? For our purposes an experiment will be any activity that generates an observable outcome.4. This chapter presents some elementary concepts of probability and statistics. Here the reader will find explanations of discrete events and their outcomes. No one ever rolls a die and discovers the outcome to be 7r for example. but one that frequently arises in financial mathematics is the binomial random variable. while the standard deviation provides a measure of the "spread" in the values the random variable can take on. A discrete outcome can take on only one of a finite number of values.Chapter 2 Discrete Probability Since the number and interactions of forces driving the values of investments are so large and complex. Some simple 15 . two important descriptive statistics will be introduced in this chapter. will be explained. 5. While statistics is a field of study unto itself. The expected value provides a number which is representative of typical values of a random variable. 2.6}. development of a deterministic mathematical model of a market is likely to be impossible. For example the outcome of a roll of a fair die can be only one of the six values in the set {1. The concept of the random variable. 3. an event is an outcome or set of outcomes of an experiment. Basic methods for determining the probabilities of outcomes will be presented.

For example the 7 could be the result of 2 on the first die and 5 on the second. rolling a pair of dice. and card draw are apparent due to the condition that there is no outcome between "heads" and "tails".If -A is an event for which P (A) — 0. An outcome of each of these experiments could be "heads". Adopting the empirical requires an investigator to conduct (or at least simulate) the exN times (where N is usually taken to be as large as practical). Events with probabilities closer to 1 are more likely to occur than events whose probabilities are closer to 0. There classical approach periment are two approaches to assigning a probability to an event. or clubs depending on the suit of the card drawn. The probability of event A will be denoted P (A). or between the two of clubs and the three of clubs respectively. . An event can also be thought of as a collection of outcomes rather than just a single outcome. The probability of an event is a real number measuring the likelihood of that event occurring as the outcome of an experiment. let the symbol A represent an event. spades. Having a flipped coin land heads up cannot be similarly decomposed. 3 . The outcome of achieving a 7 on a roll of a pair of dice could be thought of as consisting of a pair of outcomes. the example outcomes given can be thought of as "atomic" in the sense that they cannot be further broken down into simpler events. 10.16 An Undergraduate Introduction to Financial Mathematics examples of experiments include flipping a coin.1]. 7. then A is said to be an impossible event. . For the example of the coin flip or drawing a card from the deck. jack. By convention. then A is said to be a certain event. that is. queen. 0 < P (^4) < 1. king. the approach and the empirical approach. while certain events always occur. Impossible events never occur. one for each die. or ace of hearts would be a "heart" event for the experiment of drawing a card and observing its suit. diamonds. To begin the more formal study of events and probabilities. the events could be segregated into hearts. or the ace of hearts respectively. or between 6 and 7. The discreteness of a coin flip. and drawing cards from a deck. In this chapter the outcomes of experiments will be thought of as discrete in the sense that the outcomes will be from a set whose members are isolated from each other by gaps. a roll of a pair of dice. If P (A) = 1. probabilities are always real numbers in the interval [0. . . Also in this chapter the number of different outcomes of an experiment will be either finite or countable (meaning that the outcomes can be put into one-to-one correspondence with a subset of the natural numbers). Then any of the atomic events 2 . . For example the experiment of drawing a card from a standard deck.

We see this time that events A a. Since P (^4) = 1/36 and P (B) = 1/36 and the two events are mutually exclusive.P {A A B) where P (A A B) is the probability that one of the events in the overlapping non-exclusive set of . Symbolically this would be represented as P (A V B).n B are not mutually exclusive. One compound outcome of the experiment is the red coin lands on "heads" and the green coin lands on "heads" also. The classical approach is a more theoretical exercise. they cannot both simultaneously occur. or 11). Suppose instead that the investigator wants to know the probability that a total of less than 6 or an odd total results. 3. We can let event A be the outcome of a total less than 6 (that is. 9. If an investigator rolls a pair of fair dice they may want to know the probability that a total or 2 or 12 results. An investigator may wish to know the probability that A or B occurs. or 5) and let event B be the outcome of an odd total (specifically 3. assume that each outcome is equally likely. For example. The next section contains some simple rules for handling the probabilities of these compound events. 4. Then the probability of event A is estimated to be P (A) = x/N. 5. In practice the two methods closely agree. The probability of event A is then assigned the value P (A) = y/M.2 Addition Rule Suppose A and B are two events which may occur as a result of conducting an experiment. there are outcomes which overlap both events. To adjust the calculation the probabilities of the non-exclusive events should be counted only once. and then determine the number of outcomes among the total in which event A occurs (suppose this number is y). suppose a red coin and a green coin will be flipped. 7. that is. namely the odd numbers 3 and 5 are less than 6. Some experiments involve events which can be thought of as the result of two or more outcomes occurring simultaneously. the P (^4 V B) = P (A) + P (B) = 1/18. 2. a total of 2.Discrete Probability 17 During the N repetitions of the experiment the investigator counts the number of times that event A occurred. The investigator must consider the experiment carefully and determine the total number of different outcomes of the experiment (call this number M). Suppose this number is x. Let event A be the outcome of 2 and B be the outcome of 12. Thus the P(AVB) = P (A) + P (B) . especially when N is very large.

The host will then open one of the two doors not chosen in order to reveal the prize is not behind it. First. In 1990 it appeared in its present form in the "Ask Marilyn" column of Parade Magazine [vos Savant (1990)]. A contestant must guess which door hides the prize.3 Conditional Probability and Multiplication Rule During the past decade a very famous puzzle involving probability has come to be known as the "Monty Hall Problem". Determining the probability of the occurrence of A or B rests on determining the probability that both A and B occur. If A and B are mutually exclusive events then P ( A A B ) = 0 and the Addition Rule simplifies to P {A V B) = P {A) + P (B).+ 6 + 9 + 1 8 j . 2.18 An Undergraduate Introduction to Financial Mathematics outcomes occurs.1) P ( i V B)=P(A)+P(B)-P(A/\B). The host then tells the contestant they may keep their original choice or switch to the other unopened door. Theorem 2. A game show host hides a prize behind one of three doors.U + 9 J _ 11 ~ 18' Thus the calculation of the probability of event A or event B occurring is different depending on whether A and B are mutually exclusive.1 (Addition A or B occurring is Rule) For events A and B. This paradox of probability was published in a different but equivalent form in Martin Gardner's "Mathematical Games" feature of Scientific American [Gardner (1959)] in 1959 and in the American Statistician [Selvin (1975)] in 1975. the contestant announces the door they have chosen. the probability of (2. The concept outlined above is known as the Addition Rule for Probabilities and can be stated in the form of a theorem. This topic is explored in the next section. Hence P(AVB)= U+18 + 12 + 9j + U + 9. Should the contestant switch doors? .

The question "what is the probability that the second ball is green. given that the first ball was green?" could be asked. Thus the probability that both balls are green is 182/380 = 91/190. Two balls will be drawn. The reader may be asking what this situation has to do with the question originally posed. 6 of which are blue and the remaining 14 are green. When the host reveals the non-winning. Let event A be the set of outcomes in which the first ball is green and event B be the set of outcomes in which the second ball is green. it may seem that there is no advantage to switching doors. if the contestant switches they will win with probability 2/3. Numerically P (A) = 7/10 and symbolically P(AAB) Thus P (B\A) =P(AAB)/P = P(A) P{B\A) (A) = (91/190)/(7/10) = 13/19. One approach to the answer involves determining the probability that when two balls are drawn without replacement they are both green. 2. One of the classical thought experiments of discrete probability involves selecting balls from an urn. . The outcome in which both balls are green is a subset of all the outcomes in which the first ball is green. This example illustrates the concept known as conditional probability. When the contestant makes the first choice they have a 1/3 chance of being correct and a 2/3 chance of being incorrect. The probability that event A occurs given that event B occurs is denoted P (A\B). Thus the total number of outcomes is 380. The answer to this question will motivate the statement of the multiplication rule of probability. Suppose an urn contains 20 balls. the contestants first choice still has a 1/3 chance of being correct. unchosen door. Essentially the decision the contestant faces is "given that I have seen that one of the doors I did not choose is not the winning door. Consider the diagram in Fig. should I alter my choice?" The probability that one event occurs given that another event has occurred is called conditional probability. Thus the probability that both balls are green is the product of the probability that the first ball is green multiplied the probability the second ball is green. The probability that both selections are green would be the number of two green ball outcomes divided by the total number of outcomes. however. There are 20 candidates for the first ball selected and there are 19 candidates for the second ball selected.1. the second will be drawn without replacing the first. so the contestant should switch.Discrete Probability 19 At first glance when faced with two identical unopened doors. Of those outcomes (14) (13) = 182 are both green balls. but now the unchosen unopened door has a 2/3 probability of being correct.

The positive integers are placed on alternating red and black backgrounds while 0 and 00 are on green backgrounds. Then P (A A B) = . the proba- = P(A)F{B\A). (2. Theorem 2.2 can be used to find P (B\A) directly This expression is meaningful only when P (A) > 0. What is the probability that the outcome is less than 10 and more than 3 given that the outcome is an even number? Let event A be the set of outcomes in which the number is even.1 The sets of outcomes of drawing one or two balls from an urn containing blue and green balls. Let B be the set of outcomes in which the number is greater than 3 and less than 10. 2. P (A) = 10/19 if 0 and 00 are treated as even numbers.2 (Multiplication bility of A and B occurring is P(AAB) Rule) For events A and B.20 An Undergraduate Introduction to Financial Mathematics Fig. Example 2. known as the American type. The concept illustrated above is known as the Multiplication Rule for Probabilities and can be stated in the form of a theorem.2) Equation 2.1 One type of roulette wheel. has 38 potential outcomes represented by the integers 1 through 36 and two special outcomes 0 and 00.

In the election example (assuming the number of registered voters is N and that there will be no fraudulent voting) the sample space of outcomes of voter turnout is the set S = { 0 . then / maps each element in 5 to a unique real number in the interval [0. Therefore the probability that both spins will have red outcomes is P(AAB) = (9/19)(9/19) = 81/361. N}. If event A is the outcome of red on the first spin and event B is the outcome of red on the second spin. is a potential outcome of an experiment with sample space S then f(x) = P (X = x). In a more formal sense we can describe a random variable as a function which maps the set of outcomes of an experiment to some subset of the real numbers. then we as before P (A A B) = P (A) P (B\A). Symbolically a random variable for the voter turnout is the function X : S — M. In any experiment. However there is no reason to believe that the wheel somehow "remembers" the outcome of the first spin while is is being spun the second time. Thus for independent events the Multiplication Rule can be modified to P(AAB)=P{A)P (B). If S is the set outcomes of an experiment and / is the associated probability distribution function. 2 . 1 . For example the number of people who vote in an election is not known until the election is concluded. One could ask what is the probability that both spins have a red outcome.4 Random Variables and Probability Distributions The outcome of an experiment is not known until after the experiment is performed. . suppose the roulette wheel will be spun twice. if event A has no effect on event B then A and B are said to be independent. in other words f(x) is the probability that x occurs as the outcome of the experiment. The first outcome has no effect on the second outcome.1]. A probability distribution is a function which assigns a probability to each element in the sample space of outcomes of an experiment. Since a probability . .Discrete Probability 21 3/38 and p w = S = 3/20' To expand on the previous example. Often X is thought of as the > eventual numeric result of the experiment. If a. . In this situation P (B\A) = P (B). 2. .

The reader can readily determine from the table that /(2) = 3/8.3. and /(4) = 1/16. /(3) = 1/4.2 Consider a family with four children. . thus /(0) = P (X = 0) = 1/16. The gender of each child is independent from the genders of their siblings. There is one outcome in which there are no male children.1 are equally likely. N Example 2. The random variable X will represent the number of children who are male.e.4}.2. 1 B G B B B G G G B B B B G G G G Child 2 3 B B B B B G B G B B B G B G B B G G B G B G G G B G B G G G G G 4 B B B B G B B G B G G G G G B G distribution function maps an outcome to a probability then the following two characteristics are true of the function. (2) The sum of the values of the probability distribution function is unity. so assuming that the probability of a child being male is 1/2 then the 16 events shown in table 2. There are four cases in which there is a single male child and hence / ( l ) = 1/4. (1) If Xi is one of the N outcomes of an experiment then 0 < / ( X J ) < 1.1 T h e genders of four children b o r n b y t h e s a m e set of p a r e n t s . i.1. The sample space for this experiment (having four children and counting the number of boys) is the set S = {0.22 An Undergraduate Introduction to Financial Mathematics Table 2.

It will be seen to be related to the Bernoulli random variable which takes on only one of two possible values. If the number of trials is n and the probability of success on a single trial is p. It is mathematically convenient to designate the outcomes as 0 and 1. . The probability distribution function of a Bernoulli random is particularly simple.Discrete Probability 23 Several common types of random variables and their associated probability distributions will be important to the study of financial mathematics. 1 . This is the idea of the binomial random variable defined next. The number of combinations of x successes out of n trials is \x) x\(n — x)\ The probability of x successes out of n independent trials in a specified combination is. n}. A binomial random variable is parameterized by the number of repetitions of the Bernoulli experiment (referred to as trials from here on) and by the probability of success on a single trial.5 Binomial Random Variables Returning to the last example of the previous section. The gender of one child in no way influences the genders of children born before or after. then the set of possible outcomes of the binomial experiment is the set { 0 . we can think of the births of four children as four independent events. 2. . . The binomial random variable will be discussed in the next section. according to the Multiplication Rule. The probability of having a male or female child does not change between births.= . repeating a Bernoulli experiment four times. by the Addition Rule the probability of x successes out of n trials is given by the function P(X=x)= y ^ ( l -p)».3) . Since the gender of a child can take on only one of two values it is possible to think of the birth of each child as a Bernoulli event. / ( l ) = P (X = 1) = p where 0 <p< 1 and /(0) = I-p. Since the various combinations are mutually exclusive. often thought of as true or false (or sometimes as success and failure). Thus the experiment of producing four children in a family is the same as repeating the experiment of having a single child four times or. ^ ( 1 — p)n~x. A binomial random variable X is the number of successful outcomes out of n independent Bernoulli random variables. .^ _ ^ * ( i _ p r - (2.

To take an example. it can differ in at least two important ways. the expected value usually refers to the typical value of a random variable whose outcomes are not necessarily equally likely whereas the mean of a list of data treats each observation as equally likely.3 The probability that a computer memory chip is defective is 0. but not if two or more are defective. Second.02)a:(0. what would be the typical result? The notion to be explored in this section is that of expected value.044578. however.24 An Undergraduate Introduction to Financial Mathematics Thus if the probability of an individual child being born male or female is 1/2. the expected value of a random variable is the . First. The SIMM can operate correctly if one chip is defective. "if an experiment was to be performed an infinite number of times. A reader interested in a broader. The probability that a SIMM will not function is 17 p (X > 2) = J2 I /17\ J (0. In this context "statistics" refers to numbers which can be calculated from the data rather than the means and algorithms by which these numbers are calculated. if a fair die was rolled an infinite number of times. deeper. Example 2.6 Expected Value When faced with experimental data. and more rigorous background in statistics should consult one of the many textbooks devoted to the subject for example [Ross (2006)]. 2. In financial mathematics the statistical needs are somewhat more specialized than in a general purpose course in statistics.98)17-a: w 0. the probability that a family with four children will have two female children is This result agrees with the result of the more cumbersome method used to determine this probability in the previous section. summary statistics are often useful for making sense of the data. In some ways expected value is synonymous with the mean or average of a list of numerical values. Here we wish to answer the hypothetical question.02. what would be the typical outcome?" Thus we will introduce only the statistical concepts to be used later in this text. A SIMM (single in-line memory module) contains 16 chips for data storage and a 17th chip for error correction.

± + 4 . Thus we say that if F is .1 = 2.4. x (2. I + 2 .2. Since each value of X is multiplied by its corresponding probability. we will assume that the sum converges. then this sum is well-defined.4 Let random variable X represent the number of female children in a family of four children.5. The notion of the expected value of a random variable X can be extended to the expected value of a function of X. typically there are two female children and consequently two male children. P (2) = 3/8. Since X G {1.Discrete Probability 25 typical outcome of an experiment performed an infinite number of times whereas the statistical mean is calculated based on a finite collection of observations of the outcome of an experiment. Example 2.4) It is understood that the summation is taken over all values that X may assume. P (1) = 1/4. and P (4) = 1/16. we may determine this number from the formula for expected value.2. Returning to the question posed in the previous paragraph as to the typical outcome achieved when rolling a fair die an infinite number of times. 1 J ^ x=i 6 6 ^ x=i 6 2 2 Thus the average outcome of rolling a fair die is 3. what is the E [X]? The sample space of X is the set {0. 1 + 1. Assuming that births of males and females are equally likely and that all births are independent events. P (3) = 1/4. In the case that X takes on only a finite number of values with nonzero probability.3. If X is a discrete random variable with probability distribution P (X) then the expected value of X is denoted E [X] and defined as E[X] = £ ( X .3.1.4}. In families having four children.P p O ) . then the expected value of X is E[x] = ±x =1-±x=1-. 5. If X may assume an infinite number of values with probabilities greater than zero.mi= i.I + 3.6} and P (X) = 1/6 for all possible values of X. Using the binomial probability formula P (0) = 1/16. the expected value of X is a weighted average of the variable X. we have E[X] = 0.

Y) = P (X A Y). this requires the probability of two or more random variables be considered simultaneously. Suppose that the two dice can be distinguished from one another (imagine that one of them is red while the other is green). Second. However. Y) and we will understand it to mean P (X. E[aX] =aE[X}.26 An Undergraduate Introduction to Financial Mathematics a function applied to X. Thus P (1. Theorem 2. Let X be the random variable denoting the outcome of the green die while Y is a random variable denoting the outcome of the red die. then E[F(X)]=Y. namely the rolling of a pair of dice. P (X) and P (Y) refer respectively to the individual probabilities of random variable X and Y. A couple of additional comments are in order. The reader should already be familiar with one example of this situation. By the definition of expected value then E [aX] = £ daX) • P PO) =aJ2(X-P (X)) = aE [X]. joint probabilities exist even for random events which are not independent. realize that in general P (X + Y) / P (X) + P (Y). First. The joint probability function is denoted P (X. P (X + Y) refers to the probability of the sum X+Y which depends on the joint probabilities of X and Y. but is not likely to cause confusion in what follows.F(X)P(X). a Later in this work sums of random variables will become important. Thus some attention must be given to the expected value of the sum of random variables. 3) symbolizes the probability that the outcome of the red die is 1 while the outcome of the green die is 3. This is an abuse of notation.3 If X is a random variable and a is a constant. 3 ) = P ( 1 ) P ( 3 ) = 1/36 according to the multiplication rule. x When the function F is merely multiplication by a constant then the expected value takes on a simple form. If the individual dice are independent then P ( l . This naturally leads us to the issue of describing the probabilities associated with various values of the random variable X + Y. Proof. The following is true. if we wish to know the probability that the sum of the discrete random . If the experiment to be performed is rolling the pair of dice and considering the total of the upward faces then the random variable denoting the outcome of this experiment is X + Y.

Without confusion we will denote the marginal probability of X as P (X) and realize that ppO = £ p ( x .r).2) + P(3. X Y Y X An important property will be used in the proof of the next theorem. Y) < 1 for all X and Y in the discrete sample space. . The sum of the joint probability of X and Y where Y is allowed to take on each of its possible values is called the marginal probability of X.4 If Xi. Theorem 2. Returning to the dice example introduced earlier in the paragraph we see that the probability that the sum of the dice is 4 is P(X + Y = A) = Y^ X+Y=4 p X Y ( ') = P(1. x Conveniently the expected value of a sum of random variables is the sum of the expected values of the random variables. • •. For example 0 < P (X.r) = ££p(*. It is also true that ££p(x.1) _ 1 _ _ 1 + J_ _ 1 _ + 36 36 36~12' The joint probability distribution of a pair of random variables possesses many of the same properties that the probability distribution of a single random variable possesses. This notion is made more precise in the following theorem.3) + P(2.Discrete Probability 27 variables X and Y is m then by using the addition rule for probabilities P{X + Y = m) = J2 X+Y=m P X Y ( > )- The summation is taken over all combinations of X and Y such that X + Y — m.n = i.X2. Y Similarly the marginal probability of Y is denoted P (Y) and defined as P(F)=£P(X.Xk are random variables then E [Xi + X2 + • • • Xk] = E [Xi] + E [X2] + • • • + E [Xk]. . y ) .

• We can use Theorem 2. If k = 2 then E[X1+X2]= J2 X\ . Suppose n trials of a Bernoulli experiment will be conducted for which the probability of success on a single trial is 0 < p < 1. If k = 1 then the proposition is certainly true. .Xk be pairwise independent random vari- . Along the way we will also find the expected value of a Bernoulli random variable. . Theorem 2.. By assumption the trials are independent of one another and the outcomes are mutually exclusive. Later we will have need to calculate the expected value of a product of random variables. then E^ + • • • + Xk_i + Xfc] = E [Xi + • • • + Xfc_i] + E [Xk] = E [X^ + • • • + E [Xfc_i] + E [jffc] The last step is true by the induction hypothesis.*2)) X1 X2 = ^^X1P(XllX2)+^^l2P(l1)X2) X\ X2 X2 X\ X\ X2 X2 Xi = ]Tx 1 P(X 1 ) + ^ X 2 P ( X 2 ) X1 X2 = E [X^ + E [X2] • For a finite value of k > 2 the result is true by induction.28 An Undergraduate Introduction to Financial Mathematics Proof. The result of the binomial experiment can be thought of as the sum of the results of n Bernoulli experiments.5 Let Xi. X2.X2 ((X1+X2)F(X1. Random variable X represents the number of successes out of n trials.X2)) = EE«xi+x2)p(*i.4 to determine the expected value of a binomial random variable. Suppose the result is true for n < k where k > 2. This situation is not as straightforward as the case of a sum of random variables.. Let the random variable Xi be the number of successes of the ith Bernoulli trial. then E [X] = E [Xi + • • • + Xn] = E [Xi] + • • • + E [Xn] =p+---+p = np.

then E [X\ • • • Xk-iXk] = = E [X\ • • • Xk-i] E [Xk] E[X1]---E[Xk-1]E[Xk] • The last step is true by the induction hypothesis. then E [XXX2 •••Xk] = E [X^ E [X2] • • • E [X fc ]. The .X2 X1X2P(X1.Since the random variables are assumed to be independent then P {X\. Let X\ and X2 be independent random variables with joint probability distribution P (X±. Suppose the result is true for n < k where k > 2. In the next section the notions of variance and standard deviation are introduced. X2) = P {X\) P (X2). E[X1X2} = J2 X\. X2).X2) = EEM2P(X!)P(X 2 ) X\ X2 = ^X1P(X1)^X2P(X2) X\ X2 = E [XJ E [X2\ For a finite value of k > 2 the result is true by induction. They specify measures of the spread of the outcomes from the expected value. Naturally we see that when k = 1 the theorem is true.7 Variance and Standard Deviation The variance of a random variable is a measure of the spread of values of the random variable about the expected value of the random variable. 2. Next we will consider the case when k = 2. Once again we are lax in our use of notation. (2) the probability distribution of X\. there is little chance of confusion in this elementary proof. Proof.Discrete Probability 29 ables. since in the previous equation the symbol P is used in three senses ((1) the joint probability distribution of X\ and X2. and (3) the probability distribution of X2)\ however. The expected value of a random variable specifies the average outcome of an infinite number of repetitions of an experiment.

The variance may be interpreted as the average of the squared deviation of a random variable from its expected value. then the variance of X is E[X2] -E[X}2.4 and 2.2E [X] E [X] + E [X] 2 E [X 2 ] . By definition. An alternative formula for the variance is sometimes more convenient in calculations.E [ X ] ) 2 2 2 E X . we can now investigate the variance in the number of female children.E [X]2 = (0 2 )(1/16) + (l 2 )(l/4) + (2 2 )(3/8) + (3 2 )(l/4) + (4 2 )(1/16) . we should determine the variance of a Bernoulli random variable. Before investigating the variance of a binomial random variable. (2. the variance is always non-negative.6 then Var(X) = E [ X 2 ] .3 respectively. If we make use of the result of Theorem 2. • Returning to the previous example of the hypothetical family with four children. The third and fourth steps of this derivation made use of theorems 2.E [2XE [X]] + E |E [X] 2 E [X2] .6 Let X be a random variable.5). Proof.E [X] 2 .2 2 = 1.30 An Undergraduate Introduction to Financial Mathematics variance is defined as Var(X)=E[(X-E[X])2] (2. We already know that E [X] — 2. If the prob- . Var(X) = E [ ( X .2XE [X] + E [X] E [X2] . Theorem 2.5) As the reader can see from Eq. The expression X — E [X] is the signed deviation of X from its expected value.

X2.7 ables.E [Xx]] E [X2 . (2.E [X2])2 + 2(X1-E[X1])(X2-E[X2])] = E [(Xi .E [X2])] Since we are assuming that random variables X\ and X 2 are independent. Proof.E [X1])(X2 .E [X2]))2] = E [(X! . Theorem 2.5). . + p2(l-p) = p(l . then Let X\.E [X!])2] + E [(X 2 .p)(l .E [Xi]) + (X 2 .p + p) The following theorem provides an easy formula for calculating the variance of independent random variables.5 E [(X1 . then by Theorem 2.E [Xi + X 2 ]) 2 ] = E[((X1+X2)-E[X1]-E[X2])2] = E [((Xi . Take the case when k = 2. • • •. X^ be pair-wise independent random. If k = 1 then the result is trivially true. Var(X) = E [ ( X .E [X2])2] + 2E[(X1-E[X1])(X2-E[X2])] = Var (Xi) + Var (X 2 ) + 2E [(Xi . vari- Var (Xi + X 2 + • • • + Xk) = Var (X : ) + Var (X 2 ) + • • • + Var (X fc ). By the definition of variance. Var (Xi + X 2 ) = E [((Xi + X 2 ) .E [X1])(X2 .E [X!])2 + (X 2 .Discrete Probability 31 ability of success is 0 < p < 1 then according to Eq. and thus Var (Xi + X 2 ) = Var ( X ^ + Var ( X 2 ) .E [X2]] = (E[X1]-E[X1])(E[X2]-E[X2]) = 0.E [ X ] ) 2 ] = E[(X-p) 2 ] = (I-P)2P + (O-P)2(I-P) = (l-p)2p = p(l -p).E [X2])] = E [Xx .

+ Var (Xk) • Readers should think carefully about the validity of the claim that X\ — E [Xi] and X2 — E [X2] are independent in light of the assumption that X\ and X2 are independent.5 Suppose a binomial experiment is characterized by n independent repetitions of a Bernoulli trial for which the probability of success on a single trial is 0 < p < 1. Var {X) = Var ^ Xj = y i=i Var (Xj) (since trials are independent) = f>(l-p) = np(l — p) So far we have made no mention of the other topic in the heading for this section. Suppose the result has been shown true for n < k with k > 2. 2. Example 2.8 Exercises (1) Suppose the four sides of a regular tetrahedron are labeled 1 through 4. namely standard deviation. If the tetrahedron is rolled like a die. Random variable X denotes the total number of successes accrued over the n trials. Standard deviation of a random variable X is denoted by a (X) and thus a{X) = VVar(X). Then Var (X1 + • • • + Xk_! + Xk) = Var (Xi + • • • + X fc _i) + Var (Xk) = Var (Xi) + • • • + Var (Xk^) where the last equality is justified by the induction hypothesis.32 An Undergraduate Introduction to Financial Mathematics The result can be extended to any finite value of k by induction. There is little more that must be said since by definition the standard deviation is the square root of the variance. what is the probability of it landing on 3? .

As each person enters. On which draw is the card mostly likely to be the first ace drawn? (9) On the last 100 spins of an American style roulette wheel. (3) If the probability that a batter strikes out in the first inning of a baseball game is 1/3 and the probability that the batter strikes out in the fifth inning is 1/4. then what is the probability that the batter strikes out in either inning? (4) Part of a well-known puzzle involves three people entering a room. No person can see the color of their own hat. What is the probability that the first two cards will be aces? (6) Suppose cards will be drawn without replacement from a standard 52card deck.Discrete Probability 33 (2) Use the classical approach and the assumption of fair dice to find the probabilities of the outcomes obtained by rolling a pair of dice and summing the dots shown on the the upward faces. The probability that an individual receives a red hat is 1/2. Why is this a good strategy and what is the probability of winning the game? (5) Suppose cards will be drawn without replacement from a standard 52card deck. A person may decide to pass rather than to guess. the outcome has been black. one strategy the players may follow instructs a player to pass if they see the other two persons wearing mis-matched hats and to guess the opposite color if their friends are wearing matching hats. but they can see the color of the other two persons' hats. The three will split a prize if at least one person guesses the color of their own hat correctly and no one guesses incorrectly. The three people are not allowed to confer with one another once the hats have been placed on their heads. What is the probability that the fourth card drawn will be an ace given that the first three cards drawn were all aces? (8) Suppose cards will be drawn without replacement from a standard 52card deck. at random either a red or a blue hat is placed on the person's head. and the probability that the batter strikes out in both innings is 1/10. What is the probability that the second card drawn will be an ace and that the first card was not an ace? (7) Suppose cards will be drawn without replacement from a standard 52card deck. What is the probability of the outcome being black on the 101st spin? . but they are allowed to agree on a strategy prior to entering the room. At the risk of spoiling the puzzle.

10 and is zero otherwise. What is the expected value of the first ace drawn (in other words. Previously gathered evidence indicates that the defect rate for chips is 0. Typically there are nearly 105 live male births per 100 live female births. (15) Suppose a standard deck of 52 cards is well shuffled and one card at a time will be drawn without replacement from the deck. (12) Suppose that a box contains 15 black balls and 5 white balls. the outcome has been 00. . third. 2 . of the first. If two or more chips are faulty. .34 An Undergraduate Introduction to Financial Mathematics (10) On the last 5000 spins of an American style roulette wheel. . Each of the 25 chips is tested. Determine the probability distribution function for the number of black balls selected. . second. etc cards drawn. Determine the expected number of female children in a family of 6 total children using these birth ratios and ignoring infant mortality. What is the probability that a manufacturing run of chips will be discarded? (14) The probabilities of a child being born male or female are not exactly equal to 1/2. (17) For the situation described in exercise 15 determine the variance in the occurrence of the first ace drawn. . = 1. (13) Quality control for a manufacturer of integrated circuits is done by randomly selecting 25 chips from the previous days manufacturing run. Three balls will be selected without replacement from the box. (18) Show that for constants a and b and discrete random variable X that Var(aX + 6 ) = a 2 V a r ( X ) .0016. then the entire run is discarded. Find the appropriate value of the constant c. on average which will be the first ace drawn)? (16) Show that for constants a and b and discrete random variable X that E [aX + b] = aE [X] + b. What is the probability of the outcome being 00 on the 5001st spin? (11) Suppose that a random variable X has a probability distribution function / so that f(x) = c/x for a.

Continuing in this way we see that if consider only the real numbers in [0. with equal likelihood. n}. . then only the discrete set {0.1} need be considered. We will see that when many random factors and influences come together (as in the complex situation of a financial market). . the sum of the influences can be modeled by a normal random variable. continuous random variables will be described. then P (X = k/n) = ^ .1]. . If the probability of selecting either integer from this set is 1/2 then we remain in the realm of discrete probability. . a number from the interval [0. A continuous random variable can take on an infinite number of different values from a range without gaps. If we think of selecting. If we think of only the integers contained in this interval. . and distributions. 3. 1 . . 1 . 10}.1] of the form k/n where k e { 0 . . The P (X = k/10) = j ^ . A very useful type of continuous random variable for the study of financial mathematics is the normal random variable.1 Continuous Random Variables Our understanding of probability must change if we are to understand the differences between discrete and continuous random variables. .. So long as n is finite we 35 .Chapter 3 Normal Random Variables and Probability Whereas in Chapter 2 random variables could take on only a finite number of values taken from a set with gaps between the values.1] of the form k/10 where k G { 0 . Finally we will examine some stock market data and detect the presence of normal randomness in the fluctuations of stock prices. random variables. then once again our approach to probability remains discrete. in the present chapter. Calculus-based methods for determining the expected value and standard deviation of a continuous random variable will be described. Suppose we consider the interval [0.

. if X represents a continuous random variable then P (a < X < b) represents the area of the region bounded by the graph of the probability distribution function.2) These properties are the analogues of properties of discrete random variables and their distributions. (3. the x-axis. What happens as n — oo? In one sense we can think of this limiting case as the case of » continuous probability.1. and the lines x = a and x = b. By contrast P (X = x) can be greater than zero for a discrete random variable.1] becomes lim n^oc n + 1 = 0. We say continuous because in the limit. the gaps between the outcomes in the sample space disappear. of the values of the probability distribution function must be one. the likelihood of choosing a particular number from the interval [0. 3. See Fig. expressed now as an integral over the real number line. the meaningful determination is the likelihood that the random variable lies in a set.1] is 0? Surely we must choose some number. if we retain the old notion of probability. In fact. have the following property.36 An Undergraduate Introduction to Financial Mathematics are dealing with the familiar concept of discrete probability. usually an interval or finite union of intervals on the real line. However. b] the » P (a < X < b) = f f(x) dx. A random variable X has a continuous distribution if there exists a non-negative function / : M — K such that for an interval [a. This is correct but brings to mind a paradox. The sum. These notions are defined next. Interpreted graphically.1) Ja The function / which is known as the probability distribution function must. Does this imply that the probability of choosing any number from [0. (3. in addition to satisfying f(x) > 0. What is needed is a shift in our notion of probability when dealing with continuous random variables. The probabilities of individual values of the random variable are non-negative. The total area under the graph of the probability density function is unity. as explained above P (X = x) = 0 for a continuous random variable provided /(£) is continuous at t = x. oo / f(x)dx -oo = l. Instead of determining the probability that a continuous random variable equals some real number.

2) we know that oo t>b /•& -i f(x) dx = / kdx = / -oo Ja Ja b dx. Thus the probability density function for a continuous random variable on interval [a. A random variable X is uniformly distributed in the interval [a. (3. The shaded area represents the P ( a < X < b). 3.12]. Find the probability that 2 < X < 7. b] is the piecewise-defined function m TT— b—a if a < x < b — — 0 otherwise. This simplifies the evaluation of the improper integral. Suppose this constant is k. Since the length of the subinterval is proportional to the probability that X lies in the subinterval then the probability density function f{x) must be constant. b] is proportional to the length of the subinterval.1 An example of a probability distribution function. ~ a We are assuming here that the probability distribution function vanishes outside of the interval [a. The definition of the continuous uniform random variable allows us to determine the probability density function for X. Example 3. From the property expressed in Eq.Normal Random Variables and Probability 37 f (X) Fig. b) if the probability that X belongs to any subinterval of [a. Perhaps the most elementary of the continuous random variables is the uniformly distributed random variable. b].1 Random variable X is continuously randomly distributed in the interval [-1. .

Example 3.4) for the case of a discrete random variable. 75]. 3. (3. many of the proofs of results in this section will be left to the reader. (3. Using Eq.2 Expected Value of Continuous Random Variables By definition the expected value or mean of a continuous random variable with probability density function f(x) is />00 E [X] = / xf{x) dx. the probability density function of X is ^ if .3) converges.3) we have The results which follow in this section will be similar to the results that were presented for discrete random variables. In the continuous case the product of the value of the random variable and its probability density function is summed (this time through the use of a definite integral) over all values of the random variable.1 < x < 12 0 otherwise.2 Find the expected value of X if X is a continuously uniformly distributed random variable on the interval [—50. To keep the exposition as brief as possible. The expected value is only meaningful in cases in which the improper integral in (3.38 An Undergraduate Introduction to Financial Mathematics Since according to the previous discussion of uniformly distributed random variables. In most cases the notion of a discrete sum need only be replaced by a definite integral to justify these new results. The expected value of . (2.3) This equation is analogous to Eq. we have P(2<X<7) = J f(x)dx = J 2 13dx i 13 Now that continuous random variables and their probability distributions have been introduced we can begin to discuss formulae for determining their means and variances.

again. dydx+ f°° 2y . provided the integral is absolutely convergent. .y) = g|g-.Normal Random Variables and Probability 39 a function g of a continuously distributed random variable X which has probability distribution function / is denned as OO E / g(x)f(x)dx. Consequently the expected value of a function g : E 2 — R of the two > random variables X and Y is defined as oo /«oo / / -oo J — oo g(x. 2 2 . y) for which pOO OO rOO /-OO / / f{x. Joint probability distributions for continuous random variables are defined similarly to those for discrete random variables.i 3 5 /•°° 2 2 dxdy+ . A joint probability distribution for a pair of random variables is a non-negative function f(x. ^dxdy 2 .oo] x [—1. i. We can find the mean of X + Y as follows.2] whose distribution is the function f(x. Y) £ [l.y)dx dy.e. OO />00 r\ / Z / -oo J — oo o2 yoo />00 \ (x + "•*•' rt y)-1dxdy LI B Ll k L^ Lk y 3 k 1 ^^ .4) provided the improper integral converges absolutely.3 Consider the jointly distributed random variables (X. B Z 2 dxdy f°° f 2 2x .y)f(x.y)dxdy. E [<?(X)] is denned if and only if /"OO \g(x)\f(x)dx < oo. -OO (3.. Example 3.

Before turning our attention to products of continuous random variables we will define the notion of a marginal distribution. Theorem 3.2 Let X\.40 An Undergraduate Introduction to Financial Mathematics If we examine the example above we recognize that E [X + Y] = E [X] + E\Y].X2. Just as was the case for discrete random variables in the previous chapter.Xk joint probability distribution f(x\. Xk be pairwise independent random variables with joint distribution f(xi.+ E[Xk}. / -oo The marginal distribution for Y is defined similarly. the expected value of a continuous random variable can be thought of as the average value of the outcome of an infinite number of experiments. E[X1]+E[X2} + --. The variance and standard deviation of a continuous random variable again mirror the concept earlier defined for discrete random variables. X2. then E[X1X2--Xk] = E [Xi] E [X2] • • • E [Xk]. f(x. just as was the case in the previous chapter. the previous theorem is true for random variables which are dependent or independent. The additivity property of the expected value of continuous random variables is stated in the following theorem. If X and Y are continuous random variables with joint distribution f(x.y)dy. • • •. By definition two continuous random variables are independent if the joint distribution factors as the product of the marginal distributions of X and Y. %k).y) =g(x)h(y) This property is used in the proof of the following theorem. • • •.. This is true in general.1 If X\.y) then the marginal distribution for X is defined as the function oo f(x. are continuous random variables with • • •.3 Variance and Standard Deviation The variance of a continuous random variable is a measure of the spread of values of the random variable about the expected value of the random . Xfc) then E [Xi + X2 + • • • Xk] = The reader should again note that.• • . not just for the previous example. Theorem 3. 3.X2.x2.

= In the exercises the reader will be asked to prove the following two theorems which extend to continuous random variables results we have already seen for discrete random variables. if 0 < x < 1. Now by use of Theorem 3. Theorem 3. 4 Jo 3 9 9 Jo / 2x dx .4 Let X be a continuous random variable with probability distribution f(x) and let a.X2.. Theorem 3. -oo (3. • • •. [ 0 otherwise. then Var (aX + b) = a 2 Var (X).. X2. Theorem 3..b £ l .3 Let X be a random variable with probability distribution f and mean [i. Xk be pairwise independent continuous random variables with joint probability distribution f{xi. then the variance of X is E [X2~\ — /i2. = E [X] and f(x) is the probability distribution function of X. The standard deviation is sometimes .4 Find the variance of the continuous random variable whose probability distribution is given by f (x) = \ f 2a.5) where jj.Normal Random Variables and Probability 41 variable.. The variance may be interpreted as the squared deviation of a random variable from its expected value.3 we have Var {X) / J — oo x2(2x)dx1 (1)' 1 4 1 2 ~ 9 ~ 18 / 2x3 dx.- f . The variance is denned as oo / {x-fi)2f(x)dx. Example 3. Xk).. An alternative formula for the variance is given in the following theorem.5 Let X\. The reader may readily check that fi = E [X] = 2/3. then Var {Xx + X2 + • • • + Xk) = Var {Xx) + Var (X2) + • • • + Var (Xk). By definition the standard deviation of a continuous random variable is the square root of its variance.

The position of the particle after n steps will be (r — (n — r))Ax = (2r — n)Ax = mAx.42 An Undergraduate Introduction to Financial Mathematics denoted a. V N ' ' We will assume that all the steps taken by the particle are independent and identically distributed. economic. We begin by supposing that a particle sits at the origin of the x-axis and may move the left or the right a distance Ax in a time interval of length At. pg. Statisticians. Consequently the number of steps taken to the left will be n — r.m ) ) ! ' . The normal random variable can also be thought of as the limiting behavior of the binomial random variable introduced in Chapter 2. This derivation is outlined in [Bleecker and Csordas (1996). The (discrete) probability of moving left is 1/2 which also happens to be the probability of moving right. We will explore the derivation of the probability distribution for the normal random variable in this section. Let the total number of steps taken be n and the number of steps taken to the right be r. for example adult heights and weights. A continuous random variable obeying a normal distribution is frequently said to follow the "bell curve". .4 Normal Random Variables For our purposes one of the most important and useful continuous random variables will be the normally distributed random variable. 139]. The normal distribution even finds its way into the financial arena via the assumption that movements in the price of an asset are subject to a large number of incompletely understood political. 3. and social forces. thus justifying the assumption that these changes in value are related to a normal random variable (this assumption will be explored further later). Many measurable quantities found in nature seem to have normal distributions. On occasion we will denote the variance of a random variable as a2. and physical scientists frequently assume any quantity subject to a large number of small independently acting forces (regardless of their distributions) is normally distributed. We will show that if n € N and m e Z with — n < m <n and if n — m is even then the probability that the particle is at location X = mAx at time t = nAt is n!(i)n (i(n + m ) ) ! ( i ( n . mathematicians.

P (X = (m + l)Ax) 1/ ("-1)!(|) 2 ^ ( i ( n . Now suppose the claim is true for fc < n—1.( _ ! ) ) ) ! ( 2 ( I ( i + i))! ( i ( i .Normal Random Variables and Probability 43 where we have assigned m~2r-n.P (X = (m .( 1 ( 1 + ( . Therefore at time t = nAt P (X = mAx) = . This is the same probability as in Eq.3) we can state that the probability that the particle is at position mAx after n steps is P (X = mAx) = P (X = (2r n)Ax) _ .l . P(X = -Ax)P (X = Ax) 1 1U1)1 K2 2 . (2.l + m + l ) ) ! ( i ( n .1 .m ) ) ! ' .! ) ) )'! >1 ( 1 .1 + m . Using the binomial probability distribution given in Eq. This claim can also be proved by induction on n. If n = 1 then m = ±1 (the particle is not allowed to remain in place).l)Ax) + .1))! (±(n .i ) ) ! (n\( [r)\ 'iy UJ m» UJ Thus the claim is true for n = 1.2) n\ 1 r\(n --r)! ' n! (I)" (\{n + m))\ ( | ( n .2 ) ) ! ( ± ( n . (3.1)))!^/ rc-1 2 ^ ( | ( n .( m + l)))! / (n-!)!(!)" (i(n + m ))!(i( n _ m _2))! I (n-l)!(|)n (±(n + m .6).m ) ) ! » (r ' (I(„ + m ) ) ! ( i ( n .(m . If the particle will move to mAx at time nAt.m ) ) ! since n + m = 2r. then at time (n — l)At the particle must be at either (m — l)Ax or (m + l)Air.

m + 1 ) / 2 n(n+m+l)/2n(ra-m+l)/2 /2mr (n + m)( n + m + 1 )/ 2 (n .6) we obtain the following sequence of equivalent expressions.m ) ( n .7) If we apply this formula to every factorial present in expression (3. what is equivalent.m + 1 ) / 2 N (n+m+l)/2 / „ \ (n-m+l)/2 <2n-K \n + m -(n+m+l)/2 2 = ('"!) >2nix 11 + n -m/2 TON"1/2 TO 2 1 .71 + 1/2 (1 v^f(I(n + m))(^ i+l)/2 n 1+1/2 (hY (^(«-m)) (n-m+l)/2 2TT (n + m)(™+ 2 m+1 )/ 2 (n . The last l+ t/At Now we will make use of the fact that m = xjAx expression above is then equal to 2VA£ \/2?rt + xAA 2A */ xAA2A* / tA^J V~^W I xAi £Ax We will suppose the square of the step size along the x-axis is related to the step size in t via the formula (Ax) 2 = 2kAt. \ 2Ax (Ai)' .6).m ) ( n . Next we will make use of Stirling's Formula which approximates n\ when n is large. Using this relationship to replace the At's in the expression above yields Ax = 1 + —-Ax 1 •^-&x 2kt 2kt 'hnt V •Mt 2kt J V So far what we have derived is an approximation (which we will treat as an equality for large n or.44 An Undergraduate Introduction to Financial Mathematics which is the probability given in expression (3. 2Tre-nnn+1/2 2ne' -(n+m)/2(i(n+m)) (n + m + l ) / 2 (±)T t+l)/2 2^e-(n-m)/2(i(n-m))(" . To derive X . n! « V2^e~nnn+1/2 (3.1 + — n/ ( £) n 1+ -(n-m+l)/2 \ -(n+l)/2 2 /2n7T V n2 and n = t) At. for small Ax) for the probability that the particle is located at position x = mAx at time t = nAt.

t) as f(x.^o(l + aAx)l/^x = ea. . Since for \x\ > 1 it is true that 0 < e~x < e .—-Ace \ 2kt J 2 /-. Background information on finite difference approximations to derivatives can be found in [Burden and Faires (2005)]. Let us define f(x. then the integral of f(x. For k.A x 2\fkrt Ax^o V 2kt J = 1 /. t) over the entire real number line converges. X A \ 5Z^ 1 . t > 0 the expression is non-negative. We will make use of the result that limAa.Normal Random Variables and Probability 45 a probability distribution for a continuously distributed random variable we must divide the expression above by 2Ax and take the limit as Ax —> 0. Readers accustomed to numerical approximations for derivatives will recognize that we have used the centered difference formula for the first derivative in order to develop a difference quotient whose limit will be the probability distribution. X A \ 22^ / \ 1 f4-Axl I2kt J 2~7kTt^^(e_5ll)f (e"^) We must verify that this expression satisfies the properties of a probability distribution function.^ .t) = — 7 = lim 1 + ^ . suppose we write: f°° 1 0< / — J-oo 2\/kTTt e _ikt i ^ dx = S < oo. Therefore we have that I /-OC />0O Aknt J^ J^ Switching to polar coordinates by making the substitutions x = rcosO.

3) can be used to determine the average position of Fig.46 An Undergraduate Introduction to Financial Mathematics y = r cos 8. Therefore 5 = 1. (3.2. 3. The reader may consult a book on complex analysis such as [Marsden and Hoffman (1987)] for a more formal proof of this condition. and dxdy = r dr d6 produces -i /•2-7T />Z7r /'OO re ™t ikirt J0 L f° 2kt J0 re ' du M -r2'Aktdrd6 -rViktdr f Jo lim M^oo du v = lim \—e -M M^oo 1)=1. 3. Thus we have derived the probability distribution function for the particle. Notice in the third step we made use of the substitution u = r2/4kt.2 For a fixed value of t > 0 the function f(x. For a fixed value of t the probability density function has the familiar bell shape as seen in Fig. The definition of expected value given in Eq. Hence we see that the function given above satisfies the non-negativity condition and unit area condition of a probability distribution for a continuous random variable. t) resembling the often discussed "bell curve." = e 4fct will have a graph .

If the particle were initially placed at location /i but all other assumptions remained the same then the probability distribution would simply be shifted in the x direction by ji. Thus we may define the normal proba- .e . In a similar manner we may determine the variance of the position of the particle. the average location of the particle is at the origin independent of t. This is due to the assumption that the particle has no preference for movement to the left or right.3. From the derivation that a2 = 2kt we know that the "spread" in the location of the particle increases with time.Normal Random Variables and Probability 47 the particle. This can be readily seen in the surface plot shown in Fig. E[X]= / J —c -e 2Vk7Tt M 4kt dx -e 4fct im lim M / X o i 2vknt dx + lim N^oo J_N 2\/k-Kt 2 "**« dx = lim 0 Wkir ( l . as might be expected. 3.M / 4 f e t 1 + lim — \ — (e -N /4kt M-+00 V Thus. Var (X) X — CO ^ V njTTb e-^dx-(E[X})2 M 'knt 1 'knt L= lim f M^OO x2e ikt dx M 2 ikt lim 2kt Me~ M 2 / 4 f c t 0 cM ' dx 1 lim 2kt{Me~M2liktVknt M^oc 2kt ykirt Jo 2kt2Vknt 2kt J-^/Akt d x f e^'^dx e-* 2 / 4fct dx We were able to avoid evaluating the last improper integral by making use of the unit area property of the probability distribution.

9) will be frequently used in discussions involving probabilities of normal random variables.t) = retains the bell-shaped cross 2Vk-rrt section. bility distribution with mean /j. In the sequel the cumulative distribution function <f>(x) where P (X < x) 1 12-K ^ dt (3.8) When fj. and variance a2 to be the function m 1 _(*-M)2 o-Jlix (3. For example if X is a standard normal random variable then P (X < 0) = 0(0) = 1/2. Theorem 3. 3. = 0 and a = 1.48 An Undergraduate Introduction to Financial Mathematics f Fig.6 If X is a normally distributed random variable with expected value ii and variance a2. It is frequently helpful when dealing with normally distributed random variables to perform a mathematical change of variable which produces a standard normal random variable.3 For t > 0 the function f(x. this is referred to as the standard normal probability distribution. but the profile flattens and broadens as t increases. then Z = (X-n)/a is normally distributed .

7 If random variables X\.4 as n increases the histograms take on the appearance of a normal probability distribution function. . Theorem 3. The proofs of these theorems are beyond the scope of this work.20. One version of the Central Limit Theorem due to Lindeberg and Levy implies that the sample means become normally distributed as the sample size becomes large. The proof of this result is left to the reader in exercise 17. X2. If the process of collecting multiple samples and calculating their means is repeated then we can treat the sample means as random variables in their own right. Compute the means of each sample of size n and plot the frequency histogram of the means. Nevertheless we can observe the consequences of the Central Limit Theorem on data from a random number simulation. The interested reader should consult a book on the mathematical underpinnings of statistics for proofs of the various versions of the Central Limit Theorem (for example [DeGroot (1975)]). 3. As can be seen in Fig.1]. In the sequel the symbol Z will be used to denote a normally distributed random variable with standard normal distribution. Collect 5000 samples of size n (where n € {2. 3. we may collect a sample of size n and denote the mean of that sample Xn.Normal Random Variables and Probability 49 with an expected value of zero and a variance of one.5 Central Limit Theorem The name Central Limit Theorem is given to several results concerned with the distribution of sample means or the distribution of the sum of random variables.40}) of a uniformly distributed continuous random variable on the interval [0.10. • • •. The reader should keep the following points in mind. Given a random variable X which can be either discrete or continuous and which may have any probability distribution. Xn form a random sample of size n from a probability distribution with mean \i and standard deviation a then for all x n—»oo \ a J Example 3.5 The reader can replicate this example on most programmable computing devices.5. Another version of the Central Limit Theorem is due to Liapounov and concerns the asymptotic distribution of a sum of random variables.

005 0 0 0.5 we can determine that E [Yn] = 0 and Var(F„) = 1. 3.50 An Undergraduate Introduction to Financial Mathematics n=5 0. X2.02 0. . 2 .05 0.1. We will assume that for each i G { 1 .4 An illustration of the Central Limit Theorem due to Lindeberg and Levy. As the sample size increases the distribution of the sample means becomes more normal in appearance.025 0.01 0 „- • •• ' . n}. • • .07 0.01 0. Suppose the random variables X\. and 3.015 0.8 Suppose that the infinite collection {Xi}^ of random variables are pairwise independent and that for each i G N we have . .03 0.4. E [Xi] = [ii and that Var (JQ) = of.03 0.t 1/2 n= 20 1 * -" * H Fig. Theorem 3.. 06 0. • „ ' • w \ " . The following theorem establishes that as n becomes large Yn is approximately normally distributed. Now define a new random variable Y„ as Yn EILi {xt m VEr=i ai Using the assumption that the random variables are pairwise independent and Theorems 3. . . Xn are pairwise independent but not necessarily identically distributed.04 0. . 3.02 0. .

3.8 we will present a numerical simulation illustrating this result. From Eq. .Normal Random Variables and Probability 51 E [\Xi — Hi\3] < oo. the parameters [i and a are often called the drift and volatility respectively.] for i = 1. A sample for Yn of size 5000 was collected and frequency histograms were created.100] and independently selected. / / in addition.2. in place of a proof of Theorem 3. Example 3. 3.6 Lognormal Random Variables While normal random variables are central to our discussion of probability and the upcoming Black-Scholes option pricing formula. These are shown in Fig. When referring to a lognormal random variable. The mean and variance of each uniform distribution were calculated. . The reader is encouraged to pursue their own similar numerical exploration. uniformly distributed in [—100. the probability that Y < y is expressed as the integral p (Y < y)=-i= r 2 e-(*-") /2^ dL o-\/27r J-oo . oo). and a if Y = In X is a normally distributed random variable with mean fj. of equal importance will be continuous random variables which are distributed in a lognormal fashion. n where each of on and /3j were randomly. A random variable X is a lognormal random variable with parameters (j.5./3. It is readily seen that as n increases the distribution of Yn becomes more normal in appearance. and variance a2. then for any i £ l lim P (Yn <x)= <p{x) n—>oo where random variable Yn is defined as above.8) we see that for the normal random variable Y. .j3i] and Yn was computed as above. . (3. The data used in the following plots were generated by defining n continuous uniform distributions on [ai. Then a random variable Xi was randomly chosen from each uniform distribution on [ai. A lognormal random variable is a continuous random variable which takes on values in the interval (0.6 Once again.

005 — * -* • -'' 3 2 1 0 1 2 3 •-•' - 1 0 0.V*~ 3 - 2 - 1 0 1 2 .02 V 0.005 0 1 2 3 n=50 - V .)2/2a2 ^u CTV^ Jo P (X < ey).025 0. Making the change of variable t = In u we see then that P(lnX<2/) -(\nu-fj.5 An illustration of the Central Limit Theorem due to Liapounov.015 0."•'>. u Therefore a lognormally distributed random variable with parameters fi and a2 has a probability distribution function of the form 1 (<JV2-K)X /(*) (3.025 0.01 0.01 0.10) . As the sample size increases the distribution of the sum of the random variables becomes more normal in appearance. 3. i 2 3 0.52 An Undergraduate Introduction to Financial Mathematics -3 -2 -1 0 n = 20 .•••"^•s 3 Fig. - _.015 0.'\ »• •v •s.02 0.

1 a then If X is a lognormal random variable with parameters /J.13) Proof. The probability that the lognormally distributed random variable X is less than y > 0 is then P (X < y) = . By Eq. Similarly the P (X > y) = 1 — (/>(ln y) for a lognormally distributed random variable. Lemma 3.l) (3. Since X is assumed to be lognormal.3) = eM+<r 2 /2 The last equality is true since the integral represents the area under the probability distribution curve for a normal random variable with mean l-i + a2 and variance a2.4 = f ie-' 1 "*-") 2 /^ dx = d>(]ny). (3.Normal Random Variables and Probability 53 for 0 < x < oo. (TV 2lT JO X (3.12) ( > ' . . and E[X] = e ^+ CT2 / 2 Var (X) = e " 2 +ff2 (3.11) where <p(x) is the cumulative probability distribution function for a normal random variable with mean zero and variance one. Using the definitions of expected value and variance we can prove the following lemma. then Y = In X is normally distributed with E [Y] = \i and Var (Y) = a2.

206667) « 0.206667) = 1 .7 First.+a )dy_e2li+a2 ay/2TV J2(n+a') e 2(^+oe2M+^ 2 x oo / 2 „ . Example 3.+a2/2Y 2 e~(y-(fJ. what is the probability that the selling price of the stock on the next day will be higher than the price on the present day? P p i T ( n ) > l ) = P ( l n X ( n ) > 0) a = P(z> -0.581865 Thus for a security with the parameters measured and reported above. . For n > 1.e. Suppose that the selling price of a security will be measured daily and that the starting measurement will be denoted S(0). we will let S(n) denote the selling price on day n. To a good approximation the ratios of consecutive days' selling prices are lognormal random variables.+2a)y/2e2(p.( ! / . i. It is also generally assumed that the ratios are independent. If a sufficient number of measurements have been made that a financial analyst estimates the parameters of the random variable X are [i = 0. (3. there is a better than one half probability of the selling price increasing on any given day.54 An Undergraduate Introduction to Financial Mathematics Likewise by Eq.0155 and a = 0. the expressions X(n) = S(n)/S(n — 1) for n > 1 are lognormally distributed.( ^ + 2 t 7 ) ) 2 / 2 J „ _ „2 M +a dy-e ) _ g 2 M +o- (^-. then questions regarding the likelihood of future prices of the security can be answered.) D Now we will apply the concept of the lognormally distributed random variable to the situation of the price of a stock or other security.^(-0.0750.5) Var (X) = E [X2] (E[X]f i 1 r ' e "e D 2i/ p -te-M) 2 /2cT 2 dy _ feu.

In this section a theorem and several corollaries are presented which will enable the reader to calculate . To correctly approach the problem we must make use of the Central Limit Theorem. Readers can collect their own data set from many corporations' websites or from finance sites such as finaiice. 0-2(0.P(^M>1 = P((X(n))2>l) = P(21nX(n) > 0 ) _ r / .0750\/2 J = P(z> -0. P(S(B+S)/S(B)>1). we may ask what is the probability that the selling price two days hence will be higher than the present selling price? At first glance the reader may be tempted to use the fact that each observation of the random variable is independent and hence declare that the sought after probability is 0.7 Properties of Expected Value In later chapters the reader will encounter frequent references to the positive part of the difference of two quantities. X—K}. Appendix A contains a sample of closing prices for Sony Corporation stock.5622 ss 0. This is typically denoted (X-K) + = max{0. The eager reader may want to collect data for a particular stock to further test the lognormal hypothesis.61496 Thus the probability of a gain over a two-day period is nearly twice as large as may have been first suspected. This ignores the possibility that the price of the security could decrease on either (but not both) of the two days and still make a net gain over the two-day period. The assumption that ratios of selling prices sampled at regular intervals for securities are lognormally distributed will be explored further in Chapter 5. then the excess profit (if any) of the transaction is the positive part of the difference X — K.316. For instance if an investor may purchase a security worth X for price K.292271) = l-<£(-0.yahoo.0155)\ V 0.Normal Random Variables and Probability 55 Second. 3. This implies the mistaken assumption that the price of the security increased on each of the two days.292271) « 0.com.

9 Let X be a continuous random variable with probability distribution function f{x). (X — K)+ > x \i and only if X — K > x. (3.56 An Undergraduate Introduction to Financial Mathematics the expected value of (X — K)+ when X is a continuous random variable and K is a constant. which is true if and only if X > K + x. Corollary 3. °-(»-K)2/^2 V27T + fr .15) By Theorem 3. Corollary 3. P ((X . • We can immediately put Theorem 3. If K is a constant then E [(X . then E \{X . The last equality is true because f(x) is a probability distribution for the continuous random variable X.1 If X is normal random variable with mean (j.9 to work finding the expected value of the positive part of a normal random variable.9.1 is a special case of Theorem 3.14) Proof.K)+) = Proof. Theorem 3.K)cf> (!±—£) V cr J • (3.K)+] = J (J°° f(t) dt) dx. and variance a2 and K is a constant.9 the expected value of (X — K)+ is />oo -i />oo E[{X-K)+]= JK -j=^ V 27TCT Jx e-C-tf'^'dtdx . For any x > 0.K)+ >x)=l-P((X= l-P(X<K i-K+x K)+ < x) + x) f(t)dt J — CO = 1Therefore we have + E[{X-K) ]=J~(l-J_ K oo rK+x f{t) dt dx where u = t — K 1—/ f(u) du 1 dx \ f(t)dt) K dx.

the expected value becomes E[(X-K)+] K . C o r o l l a r y 3.f™ e-^2l^dt+ T JK / J-oo e-z'2dz+-j= V27rcr JK -L- r te-^l^dt {az + n)e-<' / 2 dz V 2-7T 2 J(K-n)/a ze^'2 J(K-ti)/a dz _K4> f ^ E ) + . /*00 \ E [(X .K)+] = e « + ^ ( ^ ± K According to Theorem 3.10).2 If X is a lognormally distributed random parameters \i and a2 and K > 0 is a constant then E variable with [{X .allows us to write E ax ~ K)+] = r (-±= r e z I2 dz\ dx .K<p ( ^ ^ ) (3-16) Proof.Normal Random Variables and Probability Employing integration by parts with 1 f°° u = -jL/ e~(*-M)2/2CT2 dt \J2-Ka Jx v = x 57 du = -L^e-(*-tf/*°* V27T0- dv = dx. Making the substitution z — ( l n i — j-i)/o.9 /•OO / + a ) .K)+] = J^ I J JK f(t) dt) dx t-rf/2a*dA dx \V2^O-JX t where the last equation is obtained using Eq.£ = / " e . (3.9 when X is a lognormally distributed random variable.M )/CT V27T \ a J V \ °~ / / V27T J(K-iiy/2<r2 For the purpose of completing the mean-variance analysis encountered later in Chapter 10 we must also be able to apply Theorem 3./ 2 dz+-^=r \ a ) V27T J(K.

If K is a constant then E[((X-K)+)2} ({p-2Kf + «>)t(^) + { -^^e-^.r/2^dx \/27r J(lnK-n)/a V^TTa JK \ \ 27TCT o2 J JlnK V2iraJinK V +a\-K<t> eM+^ /2 O.-.J ^ 1 ^ (»-^K ( ^ K U 3.7 31 ) .2 Let X be a normally distributed random variable with mean ji and variance a'. These results will be of use in later sections of this text. Lemma 3. (.8 Properties of Variance In this section we will derive properties of the variance for the positive part of the difference of a continuous random variable and a constant.58 An Undergraduate Introduction to Financial Mathematics This last integral will be evaluated using integration by parts with u= du = nr— I V^TT J(\nx- ~-*'<2 dz e ' dz d x v=x ^ = dx_ ' (lnx—/x)/<x _^e-0nx-M)2/2a2 '2irax Therefore we have E[(X-K)+] K r .V 2 ^ + 1 / e-0n.

18) Now we must establish a similar result for lognormally distributed random variables.^ .^ W _ ( ^ ) ) 2 - (3. Since X is normally distributed with mean \x and variance er2.-2K)2e-t2/2dt l(2K~fi)/a '2n J(2K-u)/o upon making the substitution t = (x — (/i — K))/a. °~ \2-K 1.2if) 2 + a 2 ) 0 ( ^ ^ ) \ °~ J + ( M . Corollary 3. E[((X-K)+)2} ^ 27R7 J-oo dx ((X- K) + fe-^^-K)f/2^ dx ^ / (x .Normal Random Variables and Probability 59 Proof. then X — K is normally distributed with mean \i — K and variance a2.( ^ x ^ \/2TT D Now the following result is immediately established. If K is a constant then Var ({X K)+) + J V2TT K)4> = ((/x .2„* .( ^ c .Kfe-t*-i»-K)Yl^ 2na JK 1 Z" 0 0 7= / (at + /j.^ ^ .2/2 .3 Lei X be a normally distributed random variable with mean fi and variance a2.2K)2 + a2) <p (0LZ2E\ \ & + (/i ^-^)\-^-2K)i. . J(2K-n)/<r _ 2 K ) ^ (^ _ ^ (//-2iQ<Xc_(u_2J02/2ff2 //i-2A- = ((/* . Expanding the square in the last integrand and integrating produces E[((X-K)+)2} „ m 2 l (n-2K\ V 2-K (/x 2U-2K)a J i 2 e-* 2 / 2 dt )+ _ _ _ _ ^ e f°° J(2K-u)/cr .

Corollary 3.K)+)2} = -jL/ ((i . — Proof. If K > 0 is a constant then Var ((X .4 Let X be a lognormally distributed random variable with parameters /i and a2. e-<*-°f*dz V2n is1} r°° K V2ir J (in K-u)/o g2(At+CT^/M-ln^+2 \+K2J»-\nK 2Ke^2'2Jll-lnK+ a a At last we have an expression for the variance of the positive part of the difference of a lognormal random variable and a constant. (3. If K > 0 is a constant then E[((X-K)+f] = e2^+a2)(f>{w + 2a) .20) .60 An Undergraduate Introduction to Financial Mathematics Lemma 3.K)+) = e^^l<f>(w + 2a) .19) The expression az = In x — \i will be substituted into the last integral to yield: E [{(X .-<->•*„.Kf-e-^x~^''^ 2na JK % dx dx \nK)/a.K)+)2} 1 = -±= f°° / ( e C T ^ .K)2e-z2l2 dz = -±=/ V2lT 6 (e°z+»-K)2e-z2/2dz J(\nK-tJ. e-^2^'2dz 2K ^'°'2 r J(\nK-ii)/(T f .2Ke^+a2/2(j){w + a) + K2<t>{w) where w = (fj.K)+f±e-Qnx-tf/2°a V 2iro~ Jo x L.3 Let X be a lognormally distributed random variable with parameters fi and a2.2KeIJ'+'j2/2<j>(w + a) + K2(j>(w) K<t>(w)^ (3.r \ x . E [((X .(e^+(j2l2(j){w + a)- .)/(T „2(u+<r J ) \/2lT ' J(\nK-tt)/.

Normal Random Variables and Probability 61 where w = (/x — hiK)/a.1]. (5) If X is a continuous random variable with probability density function f(x). \ 0 otherwise.1. (7) Prove Theorem 3. with probability distribution function /(*) -% if 1 < x 0 otherwise.4. (6) Prove Theorem 3.9 Exercises (1) Random variable X is continuously uniformly distributed in the interval [-4.5. oo). Determine the value of c. (8) Prove Theorem 3. P{X>a) = l-P(X<a). (11) Find the expected value and variance of the continuous random variable X whose probability distribution function is given by /(a.3. (2) Random variable X is continuously distributed on the interval (1. (9) Prove Theorem 3. (12) Show that if m and n are integers then n — m is even if and only if n + m is even. (10) Prove Theorem 3. . (3) Show using properties of the definite integral that for a continuous random variable X with probability distribution function f(x).2.) = /!Nif-1<^<2. Find P (X > 0). b e R. (4) A random variable X has a continuous Cauchy distribution with probability density function 7r(l + XZ) Show that the mean of this random variable does not exist. 3. show that E [aX + b] = aE [X] + b where a.

6. graphing calculator.01 and a = 0.4. b]. lim 2ktMe-M"'4kt (16) Using a computer algebra system. (22) Fill in the details of the proof of Corollary 3.05.2 inches. . What is the probability that the sum of the annual rainfalls in two consecutive years will exceed 30 inches? (19) The ratio of selling prices of a security on consecutive days is a lognormally distributed random variable with parameters fi = 0. (14) Use the technique of integration by parts to evaluate (15) Evaluate the following limit with k > 0 and t > 0. (23) Fill in the details of the proof of Corollary 3. What the probability of a one-day increase in the selling price? What is the probability of a one-day decrease in the selling price? What is the probability of a four-day decrease in the selling price of the security? (20) Let X be a uniformly distributed continuous random variable on the interval [a. find an expression for E \{X — K)+}. (18) The annual rainfall amount in a geographical area is normally distributed with a mean of 14 inches and a standard deviation of 3. or some numerical method evaluate the following probabilities for a standard normal random variable: (a) (b) (c) (d) P ( .3 < X < 3) P(KX<3) (17) Prove Theorem 3. If K is a constant. (21) Show that for the standard normal random variable \ — <j>(x) — <fi(—x).62 An Undergraduate Introduction to Financial Mathematics (13) Evaluate the following indefinite integral.3.K X < 1) P(-2<X<2) P ( .

For example suppose bank A issues loans at a 5% interest rate and bank B offers a savings account which pays 6% interest. it becomes possible to make a financial gain. repays the principal and interest and still has 1% of the loaned amount as profit. A person could take out a loan from bank A and place the loan into savings with bank B.1 The Concept of Arbitrage As mentioned previously one way to think about arbitrage is as the situation which arises when two financial instruments are mis-priced relative to one another." It is this absence of arbitrage which forms the basis of the derivation of the Black-Scholes equation found in Chapter 6. 4. When arbitrage opportunities arise. When it becomes time to repay bank A for the loan. bonds. The assumption that financial markets are efficient prevents such obvious arbitrage opportunities from being commonplace. Due to the mis-pricing. the person closes the savings account with bank B. or there is a betting scheme which produces a positive gain independent 63 .Chapter 4 The Arbitrage Theorem The concept known as arbitrage is subtle and can seem counter-intuitive. Suppose there is a set of possible outcomes for some experiment and that wagers can be placed on those outcomes. The Arbitrage Theorem states that either the probabilities of the outcomes are such that all bets are fair. In its basic form arbitrage exists whenever two financial instruments are mis-priced relative to one another. and stocks must be priced so as to be "arbitrage free. investors wanting to make a profit flock to the mis-priced instruments and the financial market reacts by correcting the pricing of the instruments. We will see that financial instruments such as options. In this section we will refine and make precise our definition.

Odds of the form n : m are equivalent to the odds ^ : 1. In general then if the "odds against" a particular outcome are n : 1 then the probability of the outcome is l / ( n + 1). producing a net gain of 1/2. yielding a positive payoff of 1/7. odds are used rather than probabilities.7. If the odds against an outcome are n : 1 then a unit bet will pay n units on a win and otherwise the unit bet is lost. According to the Arbitrage Theorem there is a betting strategy which generates a positive gain regardless of the outcome of the match. Suppose the "odds" of a particular sporting events outcome are quoted as "2 : 1 against. The payoffs scale multiplicatively for non-unit bets. Now suppose we wager —1 on player A and —2 on player B.2 Duality Theorem of Linear Programming In this section we will derive and discuss a result from which the Arbitrage Theorem easily follows. but do not. Since we are adopting the language of wagering. Now consider this example. Suppose the odds against player A defeating player B in a tennis match are | : 1 and the odds against player B defeating player A are | : 1. To simplify the process of placing and paying off bets. Before a statement and proof of the Arbitrage Theorem is presented a preliminary result from linear programming is needed. There is obviously something wrong with these probabilities since they should add to one. Odds against the occurrence of an event can also be expressed in the form n : m where n and m are natural numbers.64 An Undergraduate Introduction to Financial Mathematics of the outcome of the experiment. 4. though the two concepts are related by a simple formula. In this case the probability of the event occurring is m/(n + m). Converting these to probabilities we see that player A defeats player B with probability 0. Thus the probability of the desired outcome arising is 1/3.4 while player B defeats player A with probability 0. This example is hardly transparent so study of the Arbitrage Theorem would be beneficial in avoiding arbitrage opportunities in more complex financial situations. a simple example will illustrate the Arbitrage Theorem. The Duality Theorem is a familiar result to . Odds simplify the paying off of bets in the following way." We can think of this as implying that there are three outcomes to the experiment and in two of them the desired outcome does not occur and in one it does. If player A wins we will lose 3/2 on the first bet and gain 2 on the second. If player B wins we will gain 1 on the first bet and lose 6/7 on the second.

1). n and x satisfies the set of constraint equations Ax = b . 0).0. 4.0.0). If m and n are not too large this optimization task is easily accomplished. The expression x 6 R n x l is a column vector of n components.0. 0).1.1. although it is always possible to think of the cost function in the form now of 5a. X3) = (0. x2. The inner product C • X = C\X\ + C2X2 H h CnXn is referred to as the cost function. in other words at (0. A classic problem of linear programming is that of minimizing the quantity c • x subject to x being feasible. 1 + 4x2 + 8x3 + OX4. The cost function is minimized at the point (xi. The constraint in the previous example specified strict equality. then 5x\ + Ax2 + 8x3 = k defines yet another plane. Feasible sets of points in these types of optimization problems are always convex sets..1. A set is convex if for every pair of points A and B contained in the set. No modification of the cost function is necessary. the line segment connecting them also lies completely in the set. Inequality constraints can be converted to equality constraints by introducing slack variables. (1.. The expression c 6 M l x n is a row vector of n components. Thus the minimum of the cost function is 4 as can be seen in Fig. where X4 > 0 and "takes up the slack" to produce equality. For example suppose we try to minimize 5x\ + 4x2 + 8x3 subject to the constraints x\ + x2 +X3 = 1 and x is feasible. . (0.2. A is an m x n matrix while b £ R m X l is another column vector.0).2.The Arbitrage Theorem 65 people having studied linear programming and operations research.. 4. The vector x is feasible by definition if X{ > 0 for i = 1. We give a brief introduction to it here following the style of [Strang (1986)]. and (0. The constraints require the solution to the optimization problem to lie in a subset of a plane in the positive orthant of M3.. See Fig. If the minimum of the cost function is the as yet unknown value k. Some optimization problems may include inequality constraints. If the constraints of the previous example had been x\ + x2 + X3 < 1 and x feasible. 0). The cost function is minimized when the plane of the cost function first intersects the tetrahedron of feasible points. Thus x\ + x2 + x3 < 1 becomes x\ + x2 + x3 + X4 = 1.0. then the set of points where the minimum must be found would resemble a tetrahedron with vertices at (0. The smallest value of k for which the level set of the cost function intersects the constrained set of points will be the minimum of the cost function.

Similarly we will say that vector u is greater than (greater than or equal to) vector v if Ui > v\ (itj > Vi) for i = 1. . but simpler..2. . This is certainly true of the linear programming problems which are the subject of this chapter. 2 . In mathematics we frequently benefit from the ability to solve one problem by means of finding the solution to a related. For every linear programming problem of the type discussed .. . These inequalities will be denoted as appropriate u < v . We will say that vector u is less than (less than or equal to) vector v if the vectors are both elements of M lxfc and w. < Vi (u» < Vi) for i = 1. .1 The set of points in R 3 where the cost function's minimum will occur. . This new notation will simplify the statements of the theorems in this section. or u > v. 4. problem. 4.&. u > v .2.1 Dual Problems In the remainder of this chapter we will adopt a convenient extension of the notion of inequality.. u < v .66 An Undergraduate Introduction to Financial Mathematics Fig. k.

These paired optimization problems are related in the following ways.The Arbitrage Theorem 67 Fig. = b and x is feasible. .2 The minimum of the cost function is the first level set of the cost function to intersect the simplex of constrained solution points. Primal: Minimize c • x subject to Ax. there is an associated problem known as its dual. (2) the vector b moves from the constraint of the primal to the cost function of the dual. 4. Dual: Maximize y • b subject to yA < c. above. Henceforth the original problem will be known as the primal. We should note that: (1) the unknown of the dual is the row vector y £ Klx™ with n components.

n and yA < c. The non-negativity of the components of x implied by the feasibility assumption on x preserve the inequality in the latter case. the smallest value of k for which Axi + 3^2 = k intersects the set of constrained.e. Applying the level set argument as before. . . xi > 0 for i = 1 . Given a primal or a dual problem it is a routine matter to construct its partner. While both problems may seem simple. the dual is trivial. no y > 0) in the dual. Proof. • Example 4.1 (Weak Duality Theorem) 7/x andy are the unknowns of the primal and dual problems respectively. . The maximum value of 2/2 subject to the constraints must be 2/2 = 3 . = b. See Fig.68 An Undergraduate Introduction to Financial Mathematics (3) the vector c moves from the cost of the primal to the constraint of the dual. Solutions to the primal and the dual problems must satisfy the constraints Ax. .1 Consider the paired primal and dual problems: Primal: Minimize 4xi + 3^2 subject to x\ + x-i = 2 and Xi.x% > 0. We multiply the constraint in the primal by y and the constraint in the dual by x to obtain y^4x = y • b and yAx < c • x. Perhaps it is no surprise then that the solutions to the primal and dual are related. (5) there is no feasibility constraint (i.3. . 4. 7 / y b = e x then these vectors are optimal for their respective problems. then y-b < e x . feasible points for the primal is k = 6. Theorem 4. According to the Weak Duality Theorem then the minimum of the cost function of the primal must be at least 3. If y • b = c • x then y and x must be optimal since no y can make y • b larger than c • x and no x can make c • x smaller than y • b. Therefore we have y • b = yAx < c • x or naturally y • b < c • x. Dual: Maximize yi subject to yi < 3 and 2/2 < 4. (4) the constraints of the dual are inequalities and there are n of them.

yA) • x = 0. Using Eq. Thus we have proved the following theorem.3 The minimum of the cost function Ax\ +3x2 subject to the constraints xi +x% = 2 and x\. .The Arbitrage Theorem 69 constraints 0. (4. As a consequence of the Weak Duality Theorem we can determine that when x and y are optimal for their respective problems then y • b = yAx. (4. Example 4. .X2 > 0.2 We will use the previous theorem to find the optimal value of the following linear programming problem. Theorem 4. .2). .2 Optimality in the primal and dual problems requires either Xj = 0 or (yA)j = Cj for j = 1 . Primal: Minimize c • x = x\ + 2x2 + 7rc3 + 3^4 subject to Xi > 0 for .(yA) • x = 0 (c . 4. occurs at the point with coordinates (x±. n.1) we can conclude that vector x must be zero in every component for which vector c — yA is positive and vice versa.X2) = (0. Likewise by assumption in the dual yA < c which implies that each component of the vector c — yA is non-negative.5 xl Fig.1) Since x is feasible then each component of that vector is non-negative. = c • x c • x — y • Ax = 0 c • x .

By the previous theorem then the first and third components of x in the primal problem must be zero. Thus the maximum value is 19 which will also be the minimum value of the primal problem's cost function.2.4 < 3 . 4. (4.4 and Xi " 1 1-10' -2 0 1 1 X2 %3 X4 = "5" 3 Readers unfamiliar with matrix notation may wish to re-write Eq. 2/i .2 + 3 = 1 < 7.3). Therefore the primal can be recast as .4. Dual: Maximize y • b = 5j/i + 3j/2 subject to [2/1 2/2: 11-10 < [ 1 2 7 3] -20 1 1 (4. The constraints of the dual can be thought of as a system of inequalities.3) The space of points on which the dual is to be optimized exists in twodimensional space and thus is easily pictured. strict inequality is present in the first and third constraints since 2 .2(3) = . At the optimal point for the dual problem. it will be more difficult to use geometrical and graphical thinking to analyze this problem. The maximum of the cost function for the dual occurs at the point with coordinates (2/1.2/2) = (2.2) as a system of two linear equations with four unknowns. Fortunately we can also analyze this problem's associated dual problem.70 An Undergraduate Introduction to Financial Mathematics i = 1.2y2 < 1 Vi < 2 -Vi + 2/2 < 7 2/2 < 3 This solution to this set of inequalities can be pictured as the shaded region shown in Fig. Since the set of points described by the constraints exists in fourdimensional space.3.

5 Fig. Before stating and proving the Arbitrage Theorem in the next section.5. .3). The proof of this theorem may be found in [Gale et al. -15 -12.5 -10 -7. This is in fact the conclusion of the Duality Theorem. We have assumed that both the primal and dual have optimal solutions. ^ T y l = 2- . Primal: Minimize 2^2 + 3^4 subject to X{ > 0 for i = 2. Thus the Duality Theorem could be restated in the symmetric form: if either the dual or the primal has an optimal solution vector then so does the other and c • x = y • b.4) By inspection we must have x^ — 5 and thus X4 = 3. which is stated below.5 0 2. . (4. 4 and 0 "1 1-10 -20 1 1 X2 0 X4 = X2 X4 = 5 3 (4.5 -5 yi -2. (1951)]. then there is an optimal y in the dual and the minimum of c • x equals the maximum ofy-h. Theorem 4. the dual problem has a dual .4 The shaded region in the plot denotes the constrained set of points from Eq. Consequently the minimum of the cost function for the primal is seen to be 19 and it occurs at the point with coordinates (xi. 4. First. a few remarks concerning the Duality Theorem are in order.3 (Duality Theorem) If there is an optimal x in the primal.X3. I . .3) for the dual problem. . Second.0.namely the primal.The Arbitrage Theorem 71 6 4 2 -yl+y2=7 y2=3 X -2 -4 -6 .X4) = (0.X2.

Let r^ be the return for a unit bet on wager i when the outcome of the experiment is j .The Arbitrage Theorem states that the probabilities of the m outcomes of the experiment are such that the expected value of the payoff is zero.. Suppose we can place n wagers (numbered 1 through n) on the outcomes of the experiment. n..72 An Undergraduate Introduction to Financial Mathematics if the primal has an inequality constraint Ax > b in place of the equality constraint Ax = b. y[A-l] < [c0]. The primal in matrix form now resembles [A-I]\*]=h .. Dual: Maximize y • b subject to yA < c and y > 0. 4.Vm) for which m ^2yjnj=0. In conclusion the symmetric form the primal/dual pair can be stated as follows. The composite return from a betting strategy is then Y^=i xi'rij. To see this imagine introducing the slack variables xs in the primal.--. Primal: Minimize c • x subject to Ax > b and x > 0.2.xn) is called a betting strategy.. The vector x = (x\.3 The Fundamental Theorem of Finance Consider an experiment with m possible outcomes numbered 1 through m. Theorem 4. •• • ..X2. or there exists a betting strategy for which the payoff is positive regardless of the outcome of the experiment. This last inequality is equivalent to y > 0. This implies the two vector inequalities yA < c and —y < 0. then the dual possesses the extra constraint y > 0. for i = 1.3/2. while the dual becomes s . Component Xi is the amount placed on wager i.4 true: either (Arbitrage Theorem) Exactly one of the following is (1) there is a vector of probabilities y = (2/1. or .

. n + 1 satisfies the inequality constraints of the dual.. 2. . X^fci Virii = 0 for i = 1.2. .X2.1) have n + 1 components.r 2 2 • • • rn\ 1 -Tra -rim' J/1 2/2 ~r2m "0 0 0 . .. The primal problem is feasible with a minimum value of zero if and only if y is a probability vector for which all bets have an expected return of zero..m... .xn) be the betting strategy. . 2 ... n\ Y^j=i Vj = 1J a n ( i Vj > 0 iov j = 1. the dual problem is also feasible since xt = 0 for i = 1.. m. Vm 1 1 .2. This last element represents the amount of the positive payoff. • T21 . • • •.. Let vector x = (x\.." where matrix A has the form rn -T-12 • ' • 2 2 . Therefore in matrix form the primal can be thought of as minimizing 0 subject to y > 0 and r u -rl2 .. This is equivalent to Y17=i xi(~rij) + xn+i < 0 for j = 1.The Arbitrage Theorem 73 (2) there is a betting strategy x = (x\. Let the row vector c be the zero vector of m components and let the column vector b = ( 0 . Suppose the primal problem is feasible.n J-21 . xn) for which n ^2 x*rij > °> for j = 1." The unknown y is a column vector of m components.. Proof. Since c is the zero vector the cost function to be minimized is really just the constant 0.X2. In other .m. Then the maximization problem outlined above can be stated in matrix form as "Maximize x • b subject to xA < c..m.1 • • •• rnm Z/m-1 . . We would like to maximize xn+\ under the constraint that Y17=i xirij — xn+i for j = 1.. 0. .. ..r • •• -r 2 rni ~Tn2 • •• -r n 1 1 1 This is the dual of the primal problem "Minimize c • y subject to Ay = b and y > 0. According to the Duality Theorem then the maximum of the dual problem is zero which means no guaranteed profit is possible.. This matrix equation is equivalent to the system of equations. • • • ..2. 2.. . We will append another element xn+i to this vector.

3. Write down the dual problem of the following primal problem: Minimize xi + x2 + x3 subject to 2xi + x2 = 4 and X3 < 6 with Xj > 0.74 An Undergraduate Introduction to Financial Mathematics words if (1) is true then (2) is false. (2. (3) (4) (5) (6) (7) (8) .3). Now suppose the primal problem is not feasible. • 4. Minimize the cost function 7x2 + 9x3 subject to the constraints X\ + X2 + %3 = 5 and X2 + 0:3 + 2x4 = 1 with Xi > 0. (2) Sketch the region in the plane which satisfies the following inequalities: x\ + 3x2 > 6 3x\ + X2 > 6 xi > 0 X2 > 0 Is the region convex? What are the coordinates of the points at the corners of the region? Minimize the cost function x\ +X2 on the region described in exercise 2. Therefore zero is not the maximum payoff. What are A and b? Minimize the cost function x\ + X2 + 2x3 on the set where x\ + 2x2 + 3^3 = 15 and Xi > 0 for i = 1.4 Exercises (1) Suppose the odds against the three possible outcomes of an experiment are as given in the table below. (2.3).2. Introduce slack variables in the inequality constraints found in exercise 4 to produce equality constraints Ax = b . Thus we see that if (1) is false then (2) is true. According to the Duality Theorem the dual problem has no optimal solution. Outcome A B C Odds 2:1 3:1 1:1 Find a betting strategy which produces a positive net profit regardless of the outcome of the experiment. and (0.0). thus there is a betting strategy which produces a positive payoff.0). Find inequality constraints which will describe the rectangular region with corners at (0.

(10) Write down the primal companion problem to the dual problem: Maximize 2yi + 4y3 subject to y\ .y2 < 1 and y2 + 2y3 < . (11) Find the solutions to the primal and dual problems of exercise 10.The Arbitrage Theorem 75 (9) Find the solutions to the primal and dual problems of exercise 8. .1 .

This page is intentionally left blank .

they will move a unit step to the left (the negative direction). The evolution of this stochastic discrete dynamical system is called a random walk. As the step size was decreased. 5. Topics covered here could be expanded into an entire book of their own. the coin does not have to be fair). 5. A plot of the person's location versus the number of times the coin has been flipped might resemble the graph in Fig. the probability of the particle lying in a particular interval was seen to obey the normal distribution. security.Chapter 5 Random Walks and Brownian Motion In this chapter we will introduce and explain some of the concepts surrounding the probabilistic models used to capture the behavior of stock. For every time the coin lands on tails. and index prices. The bias in choosing to move left or right and the possibility of moving 77 . They will flip a coin. At first it was revealed that the location of the particle relative to the origin followed a binomial distribution. For every time the coin lands on heads.1. The person could move left or right depending on the value of a random number chosen from a normal distribution for example. Imagine a person standing at the origin of the real number line. In this chapter we will extend this discussion to a more general setting.1 Intuitive Idea of a Random Walk Earlier in Chapter 3 we analyzed the motion of a particle taking a discrete step along the rr-axis during every "tick" of the clock. they will take a unit step to the right (the positive direction). option. the main result of this chapter. Random walks do not require that the probabilities of moving left or right be equal (i. The magnitude of the movement at each stage does not have to a single unit. For our purposes we will explore just enough of the stochastic calculus to provide some justification for Ito's Lemma.e.

the probability of increase in value will be p = 1/2 and consequently the probability of decrease is the same.2 First Step Analysis A simple first application of a random walk is the situation of a person owning a share of stock whose current value is denoted S. once per day) the stock's value can increase or decrease by one unit. 5. To keep the following discussion simple. In the remainder of this chapter we will apply the concept of the random walk to modeling the movement of the value of a stock or other security. we can think of the Nth state of the random walk as a partial sum where for A" > 0.+ XN.1) . At discrete intervals (for example. (5. different distances during different steps are generalizations not present in the discussion presented in Chapter 3. This assumption implies the random walk is symmetric.1 A realization of a random walk on the real number line. If we think of Xn as a random variable which takes on value 1 with probability p = 1/2 and value — 1 with the probability. then tomorrow the value will either be S{n+1) = S(n) +1 or S(n+1) = S{n) -1.78 An Undergraduate Introduction to Financial Mathematics Fig. S(N) = S(0) + X1+X2 + --. 5. in other words if its current value on day n is S(n).

n in Eq. Proof. —n < S(n) < n. For the random walk defined in Eq.n) The last expression is the familiar probability formula for a binomial random variable with p = 1/2. If 5(0) ^ 0 then we can perform a change of variable via T(n) = S(n) — 5(0) for n = 0 . Now that we have settled on a starting state for the random walk. If we assume that the random variables X.2) Thus without loss of generality we will assume that 5(0) = 0. (5. . and define a related random walk where T(0) = 0. (1) According to Eq. (5. S(N) can be defined inductively by the formula S(N) = S(N—1)+XN for N > 0 where 5(0) is a specified initial state of the walk. S(N) — S(N—1) = XN are independent random variables.n ) = Q ( £ ) n .Random Walks and Brownian Motion 79 The random variable Xn is a generalization of the Bernoulli random variable discussed earlier. 1 . (3) P ( 5 ( n ) = TO) = ({n+nm)/2) (\)n. then P (S(ra) = 5(0) + k-(n-k)) = P (S(n) . otherwise . Lemma 5. (2) If we letTO= Ik . what states can be visited in n steps? The next lemma provides the answer and the respective probabilities of reaching these states. . (5. Likewise we see that P(T(n) = 2 f c . where k > 0 of the steps are in the positive direction and n — k > 0 are in the negative direction. The transitions in the states of the new random walk are independent. .1) with initial state (1) P (S{n) = TO) = 0 if \m\ > n.5(0) = 2k .1). contradicting the assumption that n + m i s odd. If the initial state of the random walk is 5(0) and n > 0 steps are taken. identically distributed random variables just as before. and Xj are identically distributed and independent for i ^ j then we see that the transitions between states in the random walk.1 5(0) = 0. thus if \m\ > n the partial sum cannot attain this value. (2) P (S{n) = m) = 0ifn + mis odd. (5.2) then n + m = 2k which implies n + TO is even. .

for the sake of simplicity.2 shows a random walk of a stock whose value was initially 10. In a financial setting we may think of the value of a security as following a random walk (though in reality a much more complicated one than the type we are exploring here).e. we will assume the initial state of the random walk is positive. The answers to the two primary questions will be found following the lines of reasoning laid out in [Kijima (2003)] and [Steele (2001)]. the security becomes worthless to the owner.2). (5. If the value of the security drops to 0. this is known as an absorbing boundary condition. There are two questions related to this situation that we wish to explore. i. but serves as an instructive example of the use of the concept of a random walk. (1) What is the probability that the state of the random walk crosses a threshold value of A > 0 before it hits the boundary at 0 (and hence remains there)? (2) What is the expected value of the number of steps which will elapse before the state of the random variable first crosses the A > 0 threshold? Figure 5. This situation has been so simplified and abstracted from the reality of the stock market that it is of little use in making investment decisions.80 An Undergraduate Introduction to Financial Mathematics (3) We may assume that n + m is even and thus the result is shown by Eq. S(0) > 0 and we will assume the random walk has a lower boundary of 0. Q In the previous discussion the random walk was free to move in either the positive or negative direction any integer amount as the number of steps increased. Suppose that bounds are placed in the path the particle may follow. Once again. Since the boundary at 0 is absorbing we must keep track of the smallest value which S(n) takes on. If i is the smallest non-negative integer such that rrii = 0 (and hence that S(i) = 0) then by the absorbing boundary condition S(k) = 0 for all k > i. Our attention is now focused on developing an understanding of the following conditional probability: p (S{n) = A A m„ > 0 | S(0) = i). We must further generalize our discussion of random walks to include the absorbing boundary. Thus we define m„ = mm{S(k) : 0 < k < n}. In the study of random walks. If the state of the random walk reaches the lower boundary in a finite number of steps then the state remains at that boundary value. .

A formula for this conditional probability is derived in the proof of the following lemma.. i.2 A realization of a simple random walk attempting to capture the changes in the price of a stock.2 Suppose a random walk has the form described in Eq.Random Walks and Brownian Motion 81 25 20 15 10 10 20 30 40 50 Fig.i > 0.1) in which the Xj for i = 1. In this situation P (S(n) = A | 5(0) = i) = P (S(n) = A A m „ > 0 | 5(0) = i) + P (5(n) = A A mn < 0 | 5(0) = i) by the Addition Rule (Theorem 2.1). 5. (5. then if A. are independent. By the spatial homogeneity of the .. identically distributed random variables taking on the values ± 1 with probability p = 1/2.A-i)/2J _ (5. P (S(n) = A A m„ > 0 | S(0) = i) \{n + A-i)/2) \{n. the random variable S(n) has an initial state of S'(O) = i > 0 and S(k) is allowed to wander into negative territory (and back) arbitrarily. that is..2. We will assume that A > 0 and i > 0. Lemma 5. Suppose further that the boundary at 0 is absorbing. In order to prove the lemma we will start by considering a random walk with no boundary.3) Proof. and n. This probability should depend on the three parameters A.

Finally we are able to determine . so long as \A — i\ < n and n + A — i is even (Lemma 5.82 An Undergraduate Introduction to Financial Mathematics random walk P (5(n) = A | 5(0) = i) = P (5(n) =A-i\ 5(0) = 0) {n + A-i)/2j U. 5.(n-A-i)/2j \2. provided |A + i| < n and n — A — i is even.3 any realization of a random walk which intersects the n-axis at n = i can be reflected about the axis.3 The probabilities that a symmetric random walk will follow either the solid or dashed (reflected) path are equal. P (S(n) = j | 5(0) = 0) = P (S(n) = -j | 5(0) = 0) As seen in Fig. with each path are the same.1). Therefore we have P (S(n) = A A mn < 0 | 5(0) = i) = P (S{n) = -A \ 5(0) = i) = P(S(n) = -A-i\S(0) = 0) n \(V " . 5. Now we must make use of the symmetry property of the un-restricted random walk. The probabilities associated S(n) n Fig.

In the remainder of this section we will explore the probability that CI A takes on the various natural number values given the initial state of the random walk is £(0) = i > 0.A) fi. Suppose the initial state of the random walk £(0) = i > A > 0. . .0787178.A) P (£(n . Therefore P (n 0 = n | £(0) =i-A) = p (Xn = . and the boundary at 0 is absorbing.1) = 1 A m n _ i > 0 | £(0) = i .3) questions posed. 50? Making use of the formula found in Eq. then due to the spatial homogeneity of the random walk P (£(n) = A A mn > A | £(0) =i) = P (£(n) = 0 A mn > 0 | £(0) =i-A).3) we have P (£(50) = 16 A TOSO > 0 | £(0) = 10) • 50\ _ /50 28j 1. 1 .1 A S(n . m„_i > 0 and Xn = —1. We will call the first time that the random walk £(n) equals A the stopping time. what is the probability that £(50) = 16 and S{n) > 0 for n = 0 . parameter A > 0. . f^o = n if and only if S(n — 1) = 1. . (5. Example 5. answer to the first of the two The equivalent to the now provides an (5.1) = 1 A m„_i > 0 | £(0) = i . we will define the function JAA71) \(n + A-i)/2/ (n-A-i)/2 and use it from now on. -50 Since the current notation is so bulky.12 0.1 For a symmetric random walk with initial state £(0) = 10. denoted QA.Random Walks and Brownian Motion 83 that P (S(n) = i A m n > 0 | 5(0) = i) n )(-X-( \2j \(n n A-i)/2 K{n + A-i)/2J which is previous theorem expression in Eq. Consequently P (QA = n \ £(0) = i) = \fi^i-A){n — 1) as well.(i-A)(n1).The stopping time is a random variable. .

~ 2 P(i+1) —^ 2 P(i-1)-+A +=:Pv..84 An Undergraduate Introduction to Financial Mathematics Next we take up the case in which the initial state of the random walk 0 < 5(0) = i < A. 2VA(i-l) 1 + f>A{i + l) (5. Since we know a priori that PA (0) = 0 and VA (A) = 1 then we may derive a system of linear equations from Eq.. A] from i j^ 0 to A which avoids 0.4) is the discrete difference equation approximation to the second derivative [Burden and Faires (2005)]. Therefore the probability above can be written as VA(i)= Y.-. By the Addition Rule for probability Since at each step of the random walk. It is also sometimes referred to as the discrete Laplacian equation. A — 1.• 0 0 0 0 0 0 0 0 0.. (5.• 0 0 1 VA(0) VA{1) -PA{2) VA(APA(A) .2. we can decompose the state 5(0) = i into the superposition of the two states 5(—1) = i — 1 and 5(—1) = i + 1. The analysis of this situation is modeled after the discussion in the first two chapters of [Redner (2001)].• 0 0 0 1-2 1 0 • • 0 0 0 0 1 . Since the stopping time is the first time that S(n) = A then we can think of the spatial domain of the random walk as having two absorbing boundaries: the usual one at 5 = 0 and a new one at 5 = A.. f^.1) . the increment to the left or right is independent of the previous increments.4) whose solution will give us VA{I) for i = 1. Equation (5.. The symbol Pi-^A will denote any path in the discrete interval [0.-.2 1 0 0. The symbol PPi^A will denote that probability that the random walk starting at 5(0) = i follows path Pi->A.2 1.4) 0 = VA{i ..• 1 .ZPA(i) + TA(i + 1).Finally the symbol VA{i) will denote the probability that a random walk which starts at 5(0) = i will achieve state 5 = A while avoiding the state 5 = 0. In matrix form the system of equations resembles: 1 0 0 0.

1) V0(A) 41 i 0 The expression Vo(i) is the probability that a symmetric random walk initially in state 5(0) = i achieves the state 5 = 0 while avoiding the state 5 = A. A} represent the discrete boundary set of the one-dimensional. To help in the derivation we will define the following quantities: . then if 0 < 5(0) = i < A (1) the probability that the random walk achieves state A without achieving state 0 is VA{i) = i/A. Let B = {0.. We can summarize these results in the following theorem. identically distributed random variables taking on the values ±1 with probability p = 1/2.. (2) the probability that the random walk achieves state 0 without achieving state A is Vo(i) = 1 — i/A..1 Suppose a random walk has the form described in Eq. Suppose further that the boundaries at 0 and A are absorbing. discrete random walk.Random Walks and Brownian Motion 85 The reader can verify that the solution to this matrix equation is VA(0) -PA(1) 0 1 A rA(i) PA(A-1) VA(A) = i A A-l A 1 By the symmetry of the random walk we also have Po(0) 1 ^ = W) ^o(i) Vo(A .2. are independent. Before answering the question as to the expected stopping time for the random walk attaining state A. (5. Theorem 5. we will explore the simpler issue of the stopping time of reaching either of the two absorbing boundaries.1) in which the Xj for i = 1.

( i .i ) .2nB(i) + nB(i +1). Thus it is true that fis(0) = QB(A) = 0. By the definition of expected value. Certainly it is the case that if the random walk starts on the boundary the expected stopping time is 0. where 0 < i < A and which follows path Pi^Bfis(i): the expected value of the exit time for a random walk which starts at 5(0) = i. From the Poisson equation we can derive a system of linear equations for the stopping times which are parameterized by the various initial conditions of the random walk. Equation (5.5) nB(i -1) ..5) is often called the Poisson e q u a t i o n for the stopping time.86 An Undergraduate Introduction to Financial Mathematics ujPi^B: the exit time of the random walk which starts at S(0) = i.UJ. Therefore fifl(i) E Pi i 2 Vi>- {- !] + oPP«+i>o Z_^ ^P(i+i) \u>.i ) - E Pi^B 2XP(»-D- -P 2r P(i+i)- 2 ^ ( i . The path from i — B can be decomposed > into paths from (i — 1) — B or (i + 1) — 2? with the addition of a single * > step. P(»+i)- 1] P 2 Z ^ ^ P ( i .i j + ^Wi + i) v " -' 2 1 (5. where 0 < i < A. In matrix form the system of equations has the form: 10 0 0 1-2 1 0 0 1-21 0 0 0 0 0 0 0 0 • 0 0 0" • 0 0 0 • 0 0 0 fifl(O) 0 -2 -2 fifl(l) n B (2) SIB{A-1) •1-21 • 0 0 1 ttB{A) -2 0 The matrix generated by the Poisson equation (as well as the earlier matrix generated by the Laplacian equation) are known as tridiagonal matrices . There is also a recursive relationship between the stopping times at neighboring starting points in space.

• 0 0 0 0 1 -2 1 0• • 0 0 0 -2 0 1 -2 1 • • 0 0 0 -2 (5. the first sub-diagonal. a matrix for which all entries below the main diagonal are zero. that is. Now we proceed to the second column. This technique as well as many others is explained in detail in [Golub and Van Loan (1989)]. (5. All that will be important during this process are the coefficients present in the matrix and on the right-hand side of the equation.7) by 1/2 and adding to the third row yields an augmented .6) becomes 1 0 0 0 0 -2 1 0 0 1 -2 1 0 0 0 0 " 0 0 0 -2 0 0 0 -2 (5.Random Walks and Brownian Motion 87 due to the fact that all entries in the matrices are zero except for possibly the entries on the diagonal. Thus will be work with the augmented matrix formed by appending the column vector on the righthand side of the equation to the matrix as an addition column. Gaussian elimination followed by backwards substitution.6) 0 0 0 0 0 0• • 1 -2 1 -2 0 0• • 0 0 1 0 A vertical bar is used to visually set off the augmented column from the original matrix. 1 0 0 0. and the first superdiagonal. then the system in Eq. There are many methods for solving tridiagonal linear systems. Multiplying the second row of Eq. Thus for the Poisson system we obtain the following. Gaussian elimination works by adding scalar multiples of rows of the augmented matrix to rows below in such a way as to produce an upper triangular matrix. but we will employ one of the simplest methods.7) 0 0 0 0 0 0 0 0 1 -2 1 -2 0 0 1 0 Notice that the first column of the matrix has now achieved the upper triangular form. If —1 is multiplied by the first row and then that scalar product is added to the second row. (5.

0 0 -=± I—1 0 -(»+!) 0 1 0 0 -(i + 1) Thus once the Gaussian elimination procedure is completed the augmented matrix has the upper triangular form: 10 0-2 0 0. 1 -2 1 -2 0 0 1 0 0 0 0 0 0 0 0 0 The conversion to upper triangular form will be completed by induction (see [Courant and Robbins (1969)]) on the rows.• 0 0.• 0 -A 1 0 The Gaussian elimination process is the equivalent of adding multiples of linear equations to one another.• 0 1 0• • 0 • 0 0 0 -fl0 0 0 0 0 0 0 0 0 0 0 -2 0 -3 (5.9) A-l 0.5 1x 2 0 0 0 0 " 0 0 0 -2 0 0 0 -3 (5. Suppose that Gaussian elimination has been performed on the first i rows of the augmented matrix and the ith and (i + l)th rows have the following form: 1 -2 0 0 1 0 row to The i row is multiplied by 1 — 1/i and added to the (i + 1) produce the following. If we re-write the upper triangular aug- .An Undergraduate Introduction to Financial Mathematics matrix of the form: 1 0 0 0 0 -2 1 0 0 0 .

10) The reader can understand why this method if called backwards substitution — the components of the solution are found in reverse index order. (5.i+ 3 —^^B{3)- (5. Suppose Q. The remainder of the unknowns can be found by induction.11) Thus in general QB(A—j) = j(A—j).k .• 0 A-l 1 0 nB(A.11). (5. The results are summarized in the following theorem. 2.10) produces QB(A — 1) = A — 1.1).1) VB(A) 0 Now the system is readily solved by the process of backwards substitution.2 Suppose a symmetric random walk takes place on the discrete interval {0.1) in which the Xi for i = 1. Suppose further that the boundaries at 0 and A are absorbing. • 0 0 0 0 0 0 0 0 0 0 0 OB(O) fiu(l) OB(2) 0 -2 -3 0 0 •• • 0 -— ^ .12) Example 5.1. identically distributed random variables taking on the values ±1 with probability p = 1/2.2. (5. Theorem 5.4} for which the boundaries at 0 and 4 are absorb- . If the value of SIB (j) is known then the value of QB (j — 1) can be determined from the equation J n (j-i) B J-l + nB(j) nB{j-i) =j . 1 0 0 0 •• • 0 0-2 1 0 • • 0 0 0 . (5.B(A — k) = k(A — k) is known.10) we obtain UB{A -k-l) = {k+ I) (A . are independent. then if 0 < 5(0) = i < A the random walk intersects the boundary (S = 0 or S = A) after a mean number of steps given by the formula nB(i) = i(A~i)....Random Walks and Brownian Motion mented matrix in the form of an equation we obtain the following. Setting j = A in Eq.1 i 0 0. then by setting j = A — k in Eq. (5.f 1-. A more convenient way of expressing the solution is found upon making the substitution i = A — j in Eq.2 Suppose a random walk has the form described in Eq.3. The boundary condition £IB(A) = 0 is used to seed the substitution. (5.

l)nA(i .. . This is the average number of steps required for a symmetric random walk to reach the boundary at A while avoiding the other boundary at 0. ZQ = 0 since VA(0) — 0. zA = 0 since . Now we want to determine a related quantity. . This implies the conditional exit times and path probabilities are related by the equation: nA(i)VA{i) = E Pi —A P P^A^Pi-A- (5. ^Pi — A^Pi —A Pi —A Pi —A Pi—A Note that the sums above are taken over all paths which start at 5(0) = i and exit at A and avoid 0.2 yields a formula for the mean number of steps required for a symmetric random walk to reach either boundary point of the interval [0.2vA(i)nA(i) + vA{i + i)nA(i +1) (5. The symbol QA(i) will denote the conditional exit time of the random walk in the initial state 5(0) = i which exits at boundary A while avoiding the boundary at 0. If 5(0) = i then the expected value of the stopping time of the random walk is i nB(i) 0 0 1 2 3 4 3 3 4 0 Theorem 5. The solution to this linear system will be the vector of products VA{i)SlA{i) for i = 0 .N j ^Pi — A^Pi —A / .13).13) Once again the idea of decomposing a path into a first step and the remainder of the step will be used on the right-hand side of Eq.-. To save space and simplify the notation we will define Zi = TA{i)QA(i)At the boundary where i — 0.14) Equation (5. 1 . . (5.i)nA(i .14) generates a system of linear equations which can be solved by elementary matrix algebra.1 ) .1) + \vA(i + mA{i + 1) + VA{i) -2PA(i) = vA{% . By definition this conditional exit time is / .90 An Undergraduate Introduction to Financial Mathematics ing. . SlA(i)VA(i) = /I Pi-A ( 2Pp«-1^A ^ K>(i-U-A + l \ + 2PP(i + V-A [Upii+1)-A +1]) = l-VA{i . A] based on the starting point. A. At the other boundary. . the conditional exit time.

identically distributed random variables taking on the values ± 1 with probability p = 1/2.. 2 .2 1 0 ••• 0 0 0 0 1 . Suppose further that the boundary at 0 is absorbing. . However a more useful r a n d o m variable may be one which has a normal distribution with a prescribed mean and s t a n d a r d deviation.3 8 7 3 5 0 0 Intuitive Idea of a Stochastic P r o c e s s In this section we will bridge between the discrete random walk and the continuous r a n d o m walk...1) in which the Xi for i = 1 . The random walk will stop the first time that S(n) = A. If we make use of Theorem 5. "l 0 0 0 ••• 0 0 0 1 . We now have the need for a more traditional mathematical description of these random walks.A(A) = 0. .3 Suppose a random walk has the form described in Eq.15).z\.3 Suppose a symmetric r a n d o m walk takes place on the integer-valued number line. Suppose the b o u n d a r y at 0 is absorbing. Most introductions to calculus .2 1 ••• 0 0 0 zo z\ z-i 2_ 'A 4_ A 2(A-1) A (5. An appropriate m a t h e matical model must include some random component like the flipping coin of the previous sections. . are independent...16) E x a m p l e 5.2. T h e o r e m 5.2 1 0 0 ••• 0 0 1 ZA-\ ZA 0 Finally we have established the following theorem. Then the expected value of the stopping time is fi/iW = ^ . . Let z = (zo.Random Walks and Brownian Motion 91 Q...15) 0 0 0 0 0 0 ••• 1 . A.ZA) be the solution to the matrix equation in (5. (5. A familiar place to start this investigation is the deterministic process of exponential growth and decay. fori =1. (5. If 0 < 5(0) = i < 5 then the first passage or stopping time of the state in which the r a n d o m walk attains a value of 5 is i Zi I 8 5 2 14 5 3 16 5 16 3 4 12 5 n5{i) 5.1 in order to provide expressions for VAU) we obtain the following tridiagonal linear system.

(5. the future evolution of P(t) is completely determined. (5. namely X(t + At) . (5.= .21) We write the random variable as dz rather than z since we want to think of it as the change or "delta" in a random walk. At. Expressed as an equation this statement becomes dP . If the value of P is known at a specified value of t (usually at t = 0) then this differential equation has solution P{t) = P(0)e M '. This mathematical model is described as deterministic since there is no place for random events to express themselves in the model.19) The following discussion may be easier to understand if we consider the discrete analogue of Eq. Suppose a normal random variable.20) This equation is still deterministic but becomes stochastic if we introduce a random element. see exercise 6. The constant a is frequently referred to as volatility in financial applications. The product adz{t)\/At could naturally be thought of as a normal random variable with a mean of zero and a standard deviation of a^/At. Furthermore we will assume that the dz(t) is independent of dz(s) when t T^ s. AX = nAt + <jdz(t)\TAt.92 An Undergraduate Introduction to Financial Mathematics (see for example [Stewart (1999)]) contain the mathematical model which states that the rate of change of a non-negative quantity P. (5. Once the initial value of P and the proportionality constant are set. is the proportionality constant and t usually represents time. (5. The new random part of Eq.17) where fj.18) and if we make the change of variable X = In P then this equation becomes dX = fidt.X(t) = AX = fj. The t dependency of dz comes from the need to think of z as taking on a random value at each time t. is proportional to P. (5.21) represents "noise" whose .P (5.19). dz{t) with a mean of zero and a standard deviation of one is introduced as in the following equation.17) can be rewritten as dP — =fidt. Equation (5. The proportionality constant is called a "growth rate" if [i > 0 and a "decay rate" if \i < 0.

So far in this discussion the expression VAt has not played a significant role.U] is the deterministic quantity fi(tj — tj). (5.22) are independent for tk ^ tki and identically distributed with a mean of zero. In the remainder of this section we will explore this question briefly. This might not seem so surprising if we keep in mind that the expected value of a random variable is usually a deterministic quantity.X(U) = J2(X(tk + At)-X(tk)) k=i i-i ^2 (^At + crdz{tk)\f~At\ .Random Walks and Brownian Motion 93 amplitude is a function of the volatility and the value of a normal random variable.22) Since all of the random variables {dz{tk)YkLi in Eq. This is justified in the following set of equations. For this discrete model let tj < tj = U + (j — i)At and consider the expression AX = X(tj) . Consider the cumulative change in X over several time steps each of size At.U) (5.23) Notice that the expected value of AX = X(tj)—X(ti) over the time interval [tj.i)At H(tj . then the average value of AX is fi(tj — U). (5. E[AX} = E[X(tj)-X{ti)} "i-i E ^2 (fiAt + adz(tk)\/At) ] T /iAt + ay/At ^ k=i k=i E [dz(tk)\ fi(j . but what is the purpose of the expression VAt? It will turn out that the inclusion of this expression yields a couple of simple and yet elegant results regarding continuous stochastic processes. .

Since we will generally be working with a normal random variable with mean zero and variance %/Ai we introduce the symbol dW(t) to represent it. the variance of AX would contain a nonlinear dependence on At.X{U) = fiAt + adW(U) X{ti+l) \nP(ti+1) P(ti+1) = X{U) + fiAt + adW(U) =lnP(ti)+fiAt = p(tl)eK +° M At dW + adW(ti) (5. As At — 0 this discrete model assumes the behavior of a continuous > stochastic process.01. a = 0. (5.21) can be used to generate a random walk if X(ti) is known for some value of U. / - -i) -\ _ = a (tj . (5. and P(0) = 10. In the next section we will investigate the use of this model for describing the behavior of a corporation's stock price. (5. This symbol will be used from now on.4 is the profile of end-of-day closing stock prices for a publicly traded company.21) contains VAi so that the variance of AX will be proportional to At. Without this assumption. At = 0.94 An Undergraduate Introduction to Financial Mathematics Now consider the variance of AX = X(tj) — X{ti).05.4.j — 1 are pairwise independent.. Now the purpose of the expression VAt in Eq.. Only a little imagination is needed to believe that the random walk pictured in Fig.U) (5. O'-i Var(AX) = Var I ^ \k=i (/iAt + /i-i adz(tk)V~At) cr2AtVar \k=i ^dzfa ' %) / 2 Jli a At(j J2 A .004.. 5.25) The third line in the derivation above makes use of the assignment In P = X.21) becomes apparent. 5. .24) We have assumed once again that the random variables dz(tk) for k = i. The difference Eq. In this case X{ti+1) . . then the random walk generated by the stochastic equation may resemble that shown in Fig. For example if // = 0. The random portion of the stochastic model in Eq.

05 0. 5.85 Fig.21) we must take the natural logarithm of the the stock prices. To fit the discrete form of the model used in Eq. (5.yahoo. Then we must form AX . and At. such as weekends and holidays. (5.2 9. 5. Simple descriptive statistics can be used to determine \x and a. The data is graphed in Fig.9 9. 5. since X = In P .21).4 Stock Market Example In the previous section we explored a discrete approximation to a continuous model for generating a random walk which seemed to mimic the fluctuations of the price of a corporation's stock. The model depended on several parameters which were called fi. In this section we intend to analyze the data associated with the stock price of an actual corporation and determine how to assign values to these parameters which are appropriate to that corporation.Random Walks and Brownian Motion 95 10. One convenient source of such information is the website. Appendix A contains the raw data analyzed here. The raw stock prices are playing the role of the variable P in our model.4 A realization of a random walk generated by a stochastic process of the form in Eq. a. The closing prices of the stock for 248 consecutive trading days are represented in this figure. The end-of-day closing price of a corporation's stock is generally available on the World Wide Web.com. In this section we will use data for Sony Corporation stock.5. finance.95 9. Since the data is sampled once every trading day (and days on which no trading took place. are irrelevant) then the value of At appropriate for the model to be developed is At = 1.

000555 d a y . A histogram of the values of AX calculated from the data in Appendix A appears in Fig.. 5.6. by subtracting consecutive days logarithmic prices. The mean of data is /i « —0.5 The closing prices of Sony Corporation stock between August 13. In this book we are merely taking an intuitive. despite the apparent randomness of the closing prices there is some structure and organization hidden in them. 5. .028139 day . nonrigorous approach to their use as tools for modeling prices of stocks and their derivatives. Thus. A reader interested in a more rigorous introduction to the mathematics and statistics of stochastic processes and stochastic differential equations should consult one or more of the references such as [Durrett (1996)] or [Mikosch (1998)]. The mean of this data will be our estimate of ji and the standard deviation will be an estimate of a which is also known as the volatility. If the values of AX just calculated fit the model then they will appear to be normally distributed.21) evidently matches well the behavior of the stock price of an actual company. (5. In the next section we will give some more background on stochastic processes. 2001 (day 1) and August 12. 2002 (day 248). It is certainly plausible that the values of AX summarized there are normally distributed.96 An Undergraduate Introduction to Financial Mathematics P 60 f A 3020 • 10 t 50 100 150 200 250 Fig.1 and the standard deviation is a « 0. The discrete approximation of the stochastic Eq.

. . . In particular a generalized Wiener process can be described by a straight forward extension of an ordinary differential equation. 5. t) dt + b(X. t) dW{t) (5.28) cannot be solved explicitly through integration of both sides. .05 0 0. This stochastic differential equation is obtained by adding a normally distributed random variable to the ODE.Random Walks and Brownian Motion 97 15 12.5 TJCHI -0.5 More About Stochastic Processes The equation shown in (5. dX = adt + bdW(t) (5. An Ito process has the form dX = a(X. .5 10 7.5 5 2. The random variable X can be found by integrating both sides of Eq. Jo (5.28) where the expressions a and b are each now function of time t and the random variable X.27) Other types of stochastic processes are more general still. 247.05 0. (5.1 •0.6 A histogram of the values of the In P(ti+1) . (5.21) is an example of a generalized Wiener process. 5.In P(U) for i = 1.26) to obtain X(t) = X{0) + at + f bdW(r). 2 . In general the stochastic differential equation given in Eq.1 Fig.26) The expressions a and b are constants and dW(t) is a normal random variable.

If f(x) is an (n + l)-times differentiable function on an open interval containing XQ then the function may be written as }(x) = /(*„) + /'(x„)(x . dX = adt + bdW(t) dP = aPdt + bPdW(t) First we note that while the first equation is a generalized Wiener process the second is an Ito process. In this way we will develop the fundamentally important Black-Scholes partial differential equation.I„) + ! -^-(x . This change of variable differentiation result is known as Ito's Lemma and is covered in the next section. Here we will give a brief overview of a two-variable version of Taylor's formula with remainder. What is needed is a valid procedure for changing variables in a stochastic process. 5. A standard reference containing this result is [Taylor and Mann (1983)]. if a and b represent the mean and standard deviation respectively of random variable dX (as they did for the stock price data analyzed earlier). in other words.98 An Undergraduate Introduction to Financial Mathematics Nevertheless it is an important type of equation for us to use because it will enable us to work with stochastic processes of quantities which depend in turn on other random variables having their own individual stochastic process descriptions. a stochastic calculus version of the chain rule for differentiation. In standard differential calculus if one makes the assignment X = In P then it is well understood that dX = dP/P according to the chain rule.x„f (5. Secondly.6 Ito's Lemma The proper procedure for changing variables in a stochastic differential equation can be derived using the multivariable version of Taylor's Theorem. does the second equation imply that a and b are the mean and standard deviation of the underlying variable PI The answer is no. Before we can do this we must develop an analogue of the chain rule for differentiation found in elementary calculus. Using this on a generalized Wiener process may lead us to believe that the following equations are equivalent.29) . We start with the single-variable Taylor's formula.

32) into (5. (5. Given the Ito process which describes X we will now determine the Ito process which describes Y. z0) + 2hkFyz(y0.z0) + R3(5. ZQ + k).z0) + 5 ih2Fvy(yo. The remainder term R3 contains only third order partial derivatives of F evaluated somewhere on the line connecting the points (yo.30) Using the multi-variable chain rule for derivatives we have. /'(0) = hFy(yo.z0) /"(0) = h Fyy(y0.The quantity 9 lies between x and XQ.Random Walks and Brownian Motion 99 The last term above is usually called the Taylor remainder formula and denoted Rn+i. z0) + k2Fzz(y0./(0) = F(y0 + h.29) with XQ = 0 and x = 1 to write / ( I ) = /(0) + /'(0)x + \f'\Q)x2 + R3. t) be another random variable defined as a function of X and t. Since / is a function of a single variable then we can use the single-variable form of Taylor's formula in (5.32) We have made use of the fact that Fyz = Fzy for this function under the smoothness assumptions. upon differentiating f(x) and setting x = 0. ZQ) and (yo + h. Define the function f(x) = F(yo + xh.z0). z) has partial derivatives up to order three on an open disk containing the point with coordinates (yo.This form of Taylor's will be used to derive the two-variable form.zo).34) where dW(t) is a normal random variable and a and b are functions of X and t. The other terms form a polynomial in x of degree at most n and can be used as an approximation for f(x) in a neighborhood of xo.30) we obtain A F = / ( l ) .zo) 2 + kFz(y0.zo + k) lie within the disk surrounding (yo.zo) F(y0.31) (5.z0) + 2hkFyz(y0. Thus if we substitute Eqs. .31) and (5. Suppose the function F(y. Let Y = F(X. (5.z0) + k Fzz(y0.t)dW(t) (5. 2 (5.z0 + k)= hFy(y0. ZQ + xk) where h and k are chosen small enough that (yo + h.33) This last equation can be used to derive Ito's Lemma.zo).t)dt + b(x. Let X be a random variable described by an Ito process of the form dX = a(X.z0)) + kFz(y0.

t)FxdW{t) (5. The reader is reminded that the Taylor remainder term R3 contains terms of order (At)fc where k > 2. Lemma 5. (5. 10]. Chap. Thus as At becomes small A y « Fx(aAt + bdW{t)) + Ft At + ^Fxxb2{dW(t))2. requires a more rigorous justification. The interested reader should consult [Neftci (2000). while true. Then Y is described by the following ltd process.36) where dW(t) is a normal random variable. To summarize what we have done. The dependence of a and 6 on X and t has been suppressed to simplify the notation.t). Upon simplifying.t)dt + b(X. the expression AX has been replaced by the discrete version of the Ito process. Thus we obtain the approximation A y « Fx{aAt + bdW(t)) + FtAt + ^Fxxb2At.33) we find AY = FXAX = Fx(aAt + F t At + ^FXX(AX)2 + FxtAXAt + ^Ftt{At)2 + bdW{t))2 + R3 + bdW(t)) + F t At + -Fxx{aAt + bdW{t))At + Fxt(aAt + ^Ftt(At) 2 + R3. By definition Var (dW(t)) = E [(dW(t))2] .3 (Ito's Lemma) described by the ltd process Suppose that the random variable X is dX = a(X. t))2Fxx) dt + b(X.35) The reader should be aware that this a merely an outline of a proof of Ito's Lemma.E [dW(t)]2 which implies that E [(dW{t))2] = At which we will use as an approximation to [dW(t))2.100 An Undergraduate Introduction to Financial Mathematics Using a Taylor series expansion for Y detailed in (5. t)Fx + Ft + i(6(X. dY = L(X. Suppose the random variable Y = F(X. Recall that dW{t) is a normal random variable with expected value zero and variance At. the statement of the lemma is given below. In particular the replacement of (dW(t))2 by At.t)dW(t) (5.37) .

Random

Walks and Brownian

Motion

101

Now we can return to the question raised earlier regarding the stochastic process followed by a stock price P. If X = I n P and dX = ndt + adW(t), then we can let F(X, t) = ex = P. According to Ito's Lemma then dP = i[iex + 0 + -a2ex j dt + aex dW(t) (5.38)

fi+^a2

J Pdt + aPdW{t)

Note that the coefficient of the deterministic portion of the equation contains the expression TJCT2 which would have been absent had the chain rule of deterministic calculus been used instead of Ito's Lemma. This lemma plays a role of central importance in the next chapter in which the Black-Scholes partial differential equations is derived. 5.7 Exercises

(1) Another, less formal, method for obtaining the probability that a symmetric random walk which begins in state 5(0) = i > 0 and finishes in state S = A while avoiding an absorbing boundary at 0 is to treat the Laplacian Eq. (5.4) as a continuous boundary value problem. Find all the functions f(x) which satisfy f"(x) = 0 subject to the conditions that /(0) = 0 and f(A) = 1. (2) Another, less formal, method for obtaining the stopping time for a symmetric random walk which begins in state 0 < 5(0) = i < A and finishes in either state S = 0 or S = A is to treat the Poisson Eq. (5.5) as a continuous boundary value problem. Find all the functions g(x) which satisfy g"(x) = —2 subject to the conditions that g(0) = 0 and 9(A) = 0. (3) If a stock price can change (up or down) by only one dollar per day and the probability of a unit increase is 1/2 what is the probability that the stock price will increase by $25 before decreasing by $50? (4) What is the expected number of days that will elapse before the stock described in exercise 3 increases in price by $25 or decreases by $50? (5) What is the expected number of days that will elapse before the stock described in exercise 3 increases in price by $25 while avoiding a decrease of $50?

102

An Undergraduate Introduction

to Financial

Mathematics

(6) Assuming that \x and -P(O) are constants, verify by differentiation and substitution that P(0)e' i t solves Eq. (5.17). (7) Using the growth constant and volatility of the Sony Corporation stock found earlier, use Eq. (5.25) to generate a realization of the random walk of stock prices for another year's duration. (8) Investigate the random walk of another company's stock. Collect if possible a year's worth of closing prices (P(ti), Pfo), • • •, P(tn)) for the stock of your favorite company. Plot a histogram of AX = lnP(£j_|_i) — lnP(tj) and also calculate the mean and standard deviation of this random variable. (9) Using the growth constant and volatility of the stock data you found in exercise 8, use Eq. (5.25) to generate a realization of the random walk of stock prices for another year's duration. (10) Using the chain rule for derivatives from ordinary calculus (not the stochastic version governed by Ito's Lemma) verify the expressions found for /'(0) and /"(0) in equations (5.31) and (5.32). Find an expression for the Taylor remainder P3. (11) Suppose that P is governed by the stochastic process dP = nPdt + oPdW{t)

and that Y = Pn. Find the process which governs Y. (12) Suppose that P is governed by the stochastic process dP = fiPdt + aP dW(t) and that Y = In P . Find the process which governs Y.

Chapter 6

Options

In the present world of finance there are many types of financial instruments which go by the name of options. At its simplest, an option is the right, but not the obligation, to buy or sell a security such as a stock for an agreed upon price at some time in the future. The agreed upon price for buying or selling the security is known as the strike price. Options come with a time limit at which (or prior to) they must be exercised or else they expire and become worthless. The deadline by which they must be exercised is known as the exercise time, strike time, or expiry date. In the remainder of the text the three terms will be used interchangeably. An option to buy a security in the future is called a call option. An option to sell is known as a put option. Types of options can also be distinguished by their handling of the expiry date. The European option can only be exercised at maturity, while an American option can be exercised at or before the strike time. Of the two types, the European option is simpler to treat mathematically, and will be the focus of much of the rest of this book. However, in practice, American options are more commonly traded. There is a mathematical price to be paid in terms of complexity for the added flexibility of the American-style option. Suppose stock in a certain company is selling today for $100 per share. An investor may not want to buy this stock today, but may want to own it in the future. To reduce the risk of financial loss due to a potential large increase in the price of the stock during the next three-month period, they buy a European call option on the stock with a three-month strike time and a strike price of $110. At the expiry date, if the price of the stock is above $110 and the investor now wishes to buy the stock, they have the right as holder of the call option to purchase it for $110, even if the market value of the stock is higher. Otherwise if the value of the stock is below

103

104

An Undergraduate Introduction

to Financial

Mathematics

$110 and the investor still wishes to buy the stock, they will let the call option expire without exercising it, and purchase the stock at its market price. Call options allow investors to protect themselves against paying an unexpectedly high price in the future for a stock which they are considering purchasing. We have used the language of "buying" an option. Thus options themselves have some value, just as the securities which underlie the options have a value. An important issue to consider is how are values assigned to options? In light of the Arbitrage Theorem of Chapter 4, if the option is mis-priced relative to the security, arbitrage opportunities may be created. In this chapter we will explore some of the relationships between option and security prices. We will derive the Black-Scholes partial differential equation, together with boundary and final conditions which govern the prices of European call options. During the explanations of many of the concepts in this chapter, we will refer to buying or selling a financial instrument. In the financial markets it is often possible to sell something which we do not yet own by borrowing the object from a true owner. For example investor A may borrow 100 shares of stock from investor B and sell them with the understanding that by some time in the future investor A will purchase 100 shares of the same stock and return them to investor B. Normally one buys an object first and sells it later, but in the financial arena one can often sell first and then buy later. Borrowing and selling an object with the agreement to purchase it later is called adopting a short position in the object, or sometimes shorting the object. Adopting a short position can be a profitable transaction if the investor believes the price of the object is going to decrease. Purchasing an object first and then selling it in the future is called adopting a long position.

6.1

Properties of Options

There are many relationships between the values of options and their underlying securities. The maintenance of most of these relationships is necessary to eliminate the possibility of arbitrage. We will cover some these relationships here and give the reader a taste of the way that one can prove these properties. Throughout this section we will use the following definitions. Ca: value of an American-style call option Ce: value of a European-style call option

Options

105

K: strike price of an option Pa: value of an American-style put option Pe: value of a European-style put option r: continuously compounded, risk-free interest rate S: price of a share of a security T: exercise time of an option We have used the term "risk-free" to describe the interest rate r. It is assumed that this is the rate of return of an investment which carries no risk. While the reader may wish to debate whether a complete absence of risk can ever be achieved, there are some investments which carry very little risk, for example, U.S. Treasury Bonds. Consider the cost of an American option compared with a European option. In the absence of arbitrage an American-style option must be worth at least as much as its European counterpart, i.e. Ca > Ce and Pa > Pe. Naturally we are assuming that the strike prices, strike times, and underlying securities are the same. To the contrary, suppose that Ca < Ce. Taking the position of an informed consumer, knowing that the American-style option has all the characteristics of the European option and in addition has the increased flexibility that it may be exercised early, no investor would purchase a European-style call option if it cost more than the Americanstyle option. To formulate a more mathematical proof, suppose an investor sells the European-style call option and purchases the American-style option. Since Ce — Ca > 0, the investor may purchase a risk-free bond paying interest at rate r compounded continuously. At the expiry date the bond will have value (Ce — Ca)erT. If the owner of the European option wishes to exercise the option, the investor insures this is possible by exercising their own American option. If the owner of the European option lets it expire unused, then the investor can do the same with the American option. Thus in both cases the investor makes a risk-free profit of (Ce — Ca)erT which is an example of arbitrage. It is also the case that Ce > S—Ke~rT or arbitrage is present. Assuming an absence of arbitrage and that C e < S — Ke~rT, an investor can purchase the call option and sell the security at time t = 0. Since 0 < Ke~rT < S — Ce, the investor can invest S — Ce in a risk-free bond paying interest rate r. At the strike time the bond is worth (S — Ce)erT which is greater than K. Thus if the investor chooses to exercise the option (because S at time T is greater than K), the bond can be cashed out, the security purchased for K, and there is still capital left over. In other words there is risk-free

1) To see why this relationship must be true.1) while the right-hand side is represented by portfolio B.2). the short position on the security is eliminated using the bond and purchasing the stock for S (which is less than K).1).2) An investor can borrow at interest rate r an amount equal to Pe + S — Ce. Pe + S = Ce + Ke'rT (6. The net proceeds of this transaction are the same as in the previous case. The investor must sell the security for K. This is the important Put-Call Parity Formula of Eq. At the strike time of the two options. This would allow the investor to purchase the security.106 An Undergraduate Introduction to Financial Mathematics profit. Again there will still be capital left over. The Put-Call Parity Formula implies that in an arbitrage-free setting these portfolios must have the same value. If the security is worth more than K at time T. This amount will be invested in a risk-free bond earning . the put expires worthless and the call will be exercised by its owner.3) is equivalent to the one in (6. On the other hand if S at time T is less than K. the European put option. Pe + S>Ce + Ke~rT. This generates an initial positive flow of capital in the amount of S + Pe — Ce.e. the call expires worthless and the put will be exercised by the investor. and to sell the European call option. (6. Again the investor will sell the security for K. Once again there is a risk-free profit. i. (6.4) An investor can sell the security and the European put option and buy the call option. Thus there is a risk-free profit to be realized if portfolio A is worth less than portfolio B. Suppose portfolio A is worth less than portfolio B. the investor must repay principal and interest in the amount of (Pe + S — Ce)erT. Thus the net proceeds of this transaction are K~{Pe + SCe)erT > 0 (6. i. (6. Now suppose portfolio A is worth more than portfolio B. Pe + S <Ce + Ke~rT. There is strict arbitrage-free relationship between the value of a European call and put with the same strike price and expiry date on the same underlying security. If the security is worth less than K at time T. imagine portfolio A represents the left-hand side of (6.3) since inequality (6.e.

6. Consequently there exists an arbitrage opportunity if portfolio B is worth less than portfolio A. If the security is worth less than K at time T. At that time the investor will have (S + Pe — Ce)erT. Currently an investor can purchase a . i. This type of analysis can be generalized and extended to more relevant real-world examples and thus provides a good starting point for the goal of developing option pricing formulae. After a single. indivisible unit of time T. however.6) (6.4). The analysis done here is too simple to be applied directly to any real-world example of a security. the call option expires unused and the owner of the put option will exercise it.7) While none of the formulas or inequalities developed so far enable us to price options directly. the Put-Call Parity Formula is true. Therefore the two portfolios must have the same value.2 Pricing an Option Using a Binary Model In this section we will examine a very simple option/security situation and determine the arbitrage-free price for the security. so named since the security is assumed to take on one of two values in the future. This leaves the investor with a net gain of (P e + S . The stock will be worth $50 after time T with probability p or will be worth $200 with probability 1 — p. The analysis performed here is based on a binary model. it provides an example application of the Arbitrage Theorem (Theorem 4. they are useful checks for arbitrage opportunities.e.5) since inequality (6. If the security is worth more than K at time T. Suppose the price of a share of stock is $100 currently. Ce > S .5) is equivalent to the one in (6. The investor purchases the security for K (thus canceling their short position).Ce)erT -K>0 (6. we can derive two corollary inequalities to the Put-Call Parity Formula. Thus the investor will clear their short position by buying the security for K and their net gain is as before.Ke~rT Pe > S + Ke~ rT (6. Assuming that the prices of European call and put options are nonnegative (a mild assumption).Options 107 interest at rate r until the expiry date arrives. then the put option is worthless and investor will exercise the call option. the price of the share will either be $50 or $200.4).

C(r)). At time T their net gain is -100 + 200(1 + r ) _ 1 with probability p or -100 + 50(1 + r ) _ 1 with probability 1 — p.(-100 + 200(1 + r)'1)p+ 0 == .1 0 0 ( 1 + r) + 150p + 5 0 l + 2r p= On the other hand. 0 = (-C + 50(l+r)"1)p+(-C)(l-p) 0 = SO^l + r ) " 1 -C ( -100 + 50(1 +r)-1)(l-p) C = 5° + 1 0 0 . we must calculate the present value of any potential profit. 0 =-. Suppose the investor purchases the stock initially. 3(l + r) (6. Figure 6. In the absence of arbitrage what is the value of CI The investor could buy either the option or the stock at the beginning of the time interval. Since the option or stock may be purchased at the beginning of the time interval but the profit (if any) arrives after time T has elapsed. In an arbitrage-free setting the expected value of this gain is zero. Again the expected value of this gain will be zero in the absence of arbitrage. at time T their net gain is -C + (200 . In the absence of arbitrage. (6.108 An Undergraduate Introduction to Financial Mathematics European call option whose value is C. then a risk-free profit can be made. Example 6. if the investor purchases the option initially.1 contains a parametric plot of the points (p(r).150)(1 + r ) " 1 with probability p or — C with probability 1 — p. ^ ' ~ \ $50 with probability l-p= 19/30.9) V . The arbitrage-free probability vector and the option cost both depend on the interest rate.8). If the probability p and the call option price deviate from the curve shown.9) was found using Eq. We will assume the interest rate per T unit of time is r.1 a time T = 1 Suppose the current value of the stock is 5(0) = $100 and _ J $ 2 0 0 w i t h Probability p = 11/30. the expected value of the investor's profit from either course of action should be zero regardless of which direction the price of the stock moves. The exercise time and strike price of the option are respectively T and $150. . Equation (6.

8 0. Cashing out the portfolio produces 50a.9) and (6.50y and they must pay back e 005 (100a. .50y) at time T = 1. their risk-free profit will be $64.y) = (1000.50y) One such solution is the point with coordinates (x. Liquidating the portfolio yields 200a. A risk-free profit can be generated in this situation since the option has been mis-priced according to Eqs. Therefore if the investor takes a long position in 1000 shares of the stock and shorts 3000 call options on the stock.7 0. 200a. The portfolio is originally worth 100a.-3000).+17. 6. If 5(1) = $50 the call option is not be exercised.50. > e°. Thus if x and y can be found so that the following set of linear inequalities are simultaneously satisfied then a risk free profit can be realized.6229. (6. in revenue. If 5(1) = $200 then the call option will be exercised.8). The strike price of the option is K = $150.9 1 P Fig.Options 109 22 U 20 0.50y) 50a.1 rate. + 50y in revenue. + 50y > e 005 (100a: + 17. + 17.05 (100a. Suppose an investor borrows sufficient cash to purchase x shares of stock and y call options.6 0.5 0. + 17.4 0. A parametric plot of probability and call option cost as a function of interest Suppose further that the risk-free interest rate is r = 5% and that a European call option for the stock can be purchased for C = $17.

dW(t). We begin by supposing that S is the current value of a security and that it obeys a stochastic process of the form dS Sdt + aSdW(t) (6. (6. contains the factor Fs — A. stochastic processes.AdS \ ( ^1 SFS + \a2S2Fss a2" + Ft\ + A [ [ f i + y j S ( f t + (7S(iW(t)) r ( + aS (Fs . The ideas of arbitrage.11) Suppose a portfolio of value P is created by selling the option and buying A units of the security. .3). dF = H+ SFS + \a2S2Fss + F^\ dt + aSFs.A) dW(t) S[Fs-A} + ^a2S2Fss Ft\dt (6. Under the assumption that A = -Fs.dW(t) (6. Since the portfolio is a linear combination of the the option and the security then the stochastic process governing the portfolio is dP = d(F AS) dF .12) can be simplified if we assume A = FsThis has the beneficial effect of reducing the number of stochastic terms in Eq.110 An Undergraduate Introduction to Financial Mathematics 6. Equation (6. t) is the value of any type of option (more generally called a financial derivative). Randomness is not completely eliminated since the value of the security S remains and is stochastic. the famous Black-Scholes partial differential equation.12).12) Notice the coefficient of the normal random variable. F obeys the following stochastic process.3 Black-Scholes Partial Differential Equation In this section we will derive the fundamental equation governing the pricing of options. and present value converge at this point in the study of financial mathematics. Thus the value of the portfolio is P = F — AS. then according to Ito's Lemma (Lemma 5.10) If F(S.

(6. For an introductory treatment of the heat equation see [Boyce and DiPrima (2001)]. (6. (1995)] the Black-Scholes PDE is solved by appropriate changes of variables until it becomes the heat equation.14) rF = ^a2S2Fss rF = Ft+ rSFs + ^<x2S2Fss Equation (6. The Black-Scholes PDE is a second order equation since the highest order derivative of the unknown function F present in the the equation is the second derivative. This PDE is of parabolic type.13) In an arbitrage-free setting the difference in the returns from investing in the portfolio described above or investing an equal amount of capital in a risk-free bond paying interest at rate r should be zero. which is easily solved via a technique known as separation of variables.AS) = l -a2S2Fss + Ft + Ft + rAS (6. for short. Lastly.14) is an example of a partial differential equation or P D E .12) becomes dP=(^a2S2Fss + F?jdt. The best known example of the parabolic PDE is the heat equation which is used to describe the distribution of temperature along an object. the Black-Scholes PDE is an example of a linear partial differential equation since if F\ and F^ are two solutions to the equation then c\F\ + C2-F2 is also a solution where c\ and c^ are any constants (see exercise 9). we will briefly mention a few concepts related to the BlackScholes PDE. Financial derivative products of many types obey the Black-Scholes equation.Options 111 Eq. for example a metal rod. and linearity properties. Different solutions correspond to different initial/final and . While the general theory of solving PDEs is beyond the scope of this book. 0= rPdt-dP (^a2S2Fss + Ft} dt 0 = rPdt- rP = l-a2S2FSS + Ft r(F . PDEs are often described by their order. In [Wilmott et al. Thus the following equations are true. type.

When S(T) < K the call option is said to be out of the money. The domain of the unknown function F of Eq.4 we will derive a solution to the Black-Scholes equation for the price of a European style call option. 0}. 7.14) is Q = { ( S . 6. 6.15) Since there is a final condition and no initial condition." In this section we will discuss the appropriate conditions to impose in order to obtain a solution to Eq.4 Boundary and Initial Conditions Without the imposition of side conditions in the form of initial or final conditions and boundary conditions. For a European call option we are only interested in the value of the option during the time interval [0. . (6.2. Thus the domain of the solution to Eq. Mathematicians refer to a differential equation with many possible solutions as "ill-posed. During this time interval the price of the security underlying the option may be any non-negative value. Thus the terminal value of a European call option is (S(T) . a partial differential equation can have many different solutions. Hence we can use the following equation as the final condition of the Black-Scholes PDE. When S(T) > K the call option is said to be in the money. Graphically the payoff of the portfolio is said to resemble a "hockey stick". In the next section we will describe in more detail the initial and boundary conditions relevant to the Black-Scholes PDE. where S(T) is the value of the underlying security at the exercise time and K is the strike price of the option. At time t = T the value of the security will either exceed the strike price (in which case the call option will be exercised generating an income flow of S(T) — K > 0) or the security will have a value less than or equal to the strike price (in which case the call option expires unused and has value 0).14) is a region in (S. In Sec.T].112 An Undergraduate Introduction to Financial Mathematics boundary side conditions imposed while solving the equation. We will refer to this region as f2 C (5.T)=m&x{S{T)-K.K. t)space. £)-space. F(S. (6. See Fig. where T is the exercise time of the option. (6.0} (6. i ) | 0 < S < ooandO < t < T}.14) which describes the value of a European call option. the Black-Scholes PDE is sometimes referred to as a backwards parabolic equation.K)+ = max{5(T) .

(6. S never changes and hence remains zero. t) = S. rF = Ft + rSFs + ^a2S2Fss F(S. it becomes increasingly likely that the call option will be exercised.t) = S for0<i<T. as S -> oo.e. (6. for 5* > 0. then dS = 0. namely F(0. asS-^oo.16) Now if we suppose that S is approaching infinity. F(0. the call option would never be exercised and hence must have zero value. . S will exceed > any finite value of K. The state S — 0 is an absorbing boundary condition for the random walk of S. i. the difference S — K fa S and hence > we have the remaining boundary condition F(S. Also as S — oo.t) = 0 F{S.K)+ for (S. We see from Eq. 6.i) = 0.2 The piecewise linear curve representing the payoff of a European call option. Thus we have derived one boundary condition.Options 113 (S(T)-K)+ S(T) Fig.17) Thus to summarize the Black-Scholes equation and its final and boundary conditions for a European call option we have the following set of equations.10) that if S = 0. T) = (S(T) . Thus on the portion of the boundary of ft where 5 = 0. t) in ft. (6. since as S — oo.

(2) Consider a European call option with a strike price of $60 which costs $10. (4) What is the minimum price of a European style call option with an exercise time of three months and a strike price of $26 for a security whose current value is $29 while the continuously compounded interest rate is 6%? (5) If a share of a security is currently selling for $31.100]. exercise time. and the risk-free interest rate is 10%.114 An Undergraduate Introduction to Financial Mathematics 6.5). and strike price for both options are the same. Draw a graph illustrating the net payoff of the option for stock prices in the interval [0. (10) What final and boundary conditions are appropriate for the BlackScholes PDE (6. what is the arbitrage-free strike price for the security? (7) What is the minimum price of a European put option with an exercise time of two months and a strike price of $14 for a stock whose value is $11 while the continuously compounded interest rate is 7%? (8) If a share of a security is currently selling for $31. a three-month European call option is $3 with a strike price of $31. (7.14) when F represents a European-style put option? .t) also solves the Black-Scholes equation.25.t) = c\fi(S. Show that if c\ and ci are constants then the function f(S. a three-month European put option is $1. t) each solve the Black-Scholes Eq. t) and /2(5. a three-month European call option is $3. and the risk-free interest rate is 10%. the strike price for both options is $30.t) + C2f2(S. what is the arbitrage-free European put option price for the security? (6) If a share of a security is currently selling for $31. a three-month European put option is $2.5 Exercises (1) Show that in the absence of arbitrage Pa > Pe where the underlying security. what arbitrage opportunities are open to investors? (9) Suppose the functions fi(S. a three-month European call option is $3. (3) Show that in the absence of arbitrage the value of a call option never exceeds the value of the underlying security. and the risk-free interest rate is 10%.

1) . and that The Fourier Transform of / is denoted T{f{x)} or often as f(w). In the present chapter we will solve the Black-Scholes equation with boundary and final conditions appropriate for a European style call option.Chapter 7 Solution of the Black-Scholes Equation In Chapter 6 the Black-Scholes partial differential equation was derived and summarized in Eq.14). In [Wilmott et al. a brief introduction to the relevant operations are included in this chapter. 7.1 Fourier Transforms Suppose the function f(x) is defined for all real numbers. Several methods are available to solve this equation. Every European style option satisfies this PDE. (6. By definition the Fourier Transform of / is oo / 115 f(x)e-*wxdx. Their change of variable approach will be mimicked here. The differences in the options are due to different boundary and final conditions. (1995)] the Black-Scholes PDE is transformed through a sequence of changes of variable to an ordinary differential equation which is solved by elementary means. Some take the limit of a discrete time binomial model of security prices (see [Ross (1999)]). but the Fourier Transform will be used to solve the equation. A more complete introduction to the Fourier Transform can be found in either [Greenberg (1998)] or [Jeffrey (2002)]. The Fourier Transform is a standard mathematical device used to convert certain types of partial differential equations into ordinary differential equations. -OO (7. For readers unfamiliar with the Fourier Transform.

9. The Euler Identity. POO /H= / / . It is possible to state very general but technical conditions in such an existence theorem. e10 = cos 8 + i sin 9 was also used to simplify the result [Churchill et.1 of [Greenberg (1998)]. M) for arbitrary M > 0. The symbol i = \/—T. f(x)e-iwxdx = — . / ( s ) = Its Fourier Transform is readily found through the following calculation. al. For our purposes we will adopt a less technical theorem which states that if the domain of / is all real numbers and if • / and / ' are piecewise continuous on every interval of the form [-M. The Fourier Transform also has an interesting effect on derivatives of functions. and oo / \f(x)\dx -oo converges.(e-iw iw 2 = — sin w w -eiw) In the previous example we carried out integration of a complex exponential as if it followed the same rules of integration as real-values functions. There are many theorems in the literature on the Fourier Transform which list conditions under which the Fourier Transform will exist. then the Fourier Transform of / exists according to Theorem 17. Example 7. Suppose the Fourier Transform of / exists and that / is differ- . The Fourier Transform maps the original function / which is a function of a. (1976)].1 Consider the piecewise-defined function /lifM<l [ 0 otherwise. to a new function / which depends on w which is often thought of as a frequency. This is in fact true.116 An Undergraduate Introduction to Financial Mathematics provided the improper integral converges.

2) for the case where n = 1. (7. merely the integration of the . . The technique of proof by induction is used to establish this result for all higher order derivatives. Theorem 7. . Since / vanishes as \x\ — oo the > leading integral is zero. The result has already been shown in Eq. The convolution involves no complex integration. . f^n~1\x) are all Fourier transformable {n) (n) and if f {x) exists (where neN) then F{f (x)} = {iw)nf{w). By definition the Fourier Convolution of two functions / and g is OO / f(x-z)g(z)dz. F{f(k+l\x)}= J —OO f{k+1) (x)e~iwx dx oo r°° = f(k\x)e-iwx OO -OO - f(k\x)(-iw)e-iwxdx f(k){x)e f^(x)e-lwxdx iw / / -OO = iw(iw)kf{w) = (iw)k+1f(w) The induction assumption was used in the second to the last line of this equation. • Another Fourier Transform result which we will have need for later concerns the transform of the convolution of two functions. -oo (7. .Solution of the Black-Scholes Equation 117 entiable for all real numbers.1 If f(x).2) The technique of integration by parts was used to shift the derivative off of / and onto the exponential function.3) The Fourier Convolution is not the same as the Fourier Transform. Suppose the result has been proved for n = k. f'(x). Proof. This property of the Fourier Transform can be extended to higher order derivatives. oo / -OO OO / OO f(x)(-iw)e -oo / -OO f(x)e~iwx dx = iwf(w) (7.

2 Inverse Fourier Transforms So far we have seen that the Fourier Transform can be used to map functions of x. in other words the Fourier Transform of the Fourier Convolution of f and g is the product of the Fourier Transforms of f and g. any useful solution to the Black-Scholes equation . / / f(x-z)g(z)e-iwxdz dx f(x .z)g(z)e~ lwxdx dz . oo r />oo / J—oo dz g(z) / /(«)e-™<» '>d U du dz f(u)e~ •IWZ dz + 00 J-oo g(z)e l_«^ — oo g{z)e-™Af ) oo \_J — / g(z)e -oo = f{w)g{w) U 7.J — oo oo !_•/ —oo oo oo r /»oo ~iwx dx dz g{z) / f(x . We intend to apply this transformation to the Black-Scholes partial differential equation and its associated side conditions.z)e J— oo Since the order of integration-oo a multiple integral is irrelevant (only the in region over which the integration takes place is meaningful). However. their derivatives. It is possible to calculate the Fourier Transform of this convolution. equality is preserved above.118 An Undergraduate Introduction to Financial Mathematics product of two real-valued functions. Now the change of variable x = u + z is used to rewrite the interior integral above. We begin by applying the definitions of Fourier Transform and Convolution. Theorem 7.2 F{(f * g)(x)} = f{w)g(w). oo r />oo / / oo oo " /»O0 f{x-z)g{z)dz ' dx J. Proof. and their convolutions to other functions which depend on a frequency-like variable w.

It is necessary to have a method for transforming a function of w back into a function of x.6) (7.7) for S € (0. though they are all equivalent and compatible with the versions given above in Eqs. 7.2 Suppose a is a positive constant. T). + + rSFs for t € [0. d Readers with some prior exposure to the Fourier Transform may prefer more symmetric definitions of the transform and the inverse transform. (7. This is the purpose of the Inverse Fourier Transform.1) and (7. oo) (7.5) (7. S G (0. T). oo) as S -+ oo for t G [0. By definition the inverse Fourier Transform of f(w) given by ^HfM} = ^J fMelwx dw. t) = 0. we have the following three equations rF = Ft + -a2S2Fss F(S.2) 1 2ix{a — ix) / ^ " J0O 2TT J (-a+ix)«. An infinite number of variations on the formulas can be stated.S) F(0. then the inverse Fourier Transform of e _a '' ! "' is 1 f°° T-lre-a\wh = _ 1 _ / oo -a\w\ iwxd -i /•oo 1 /* 2?r 0 /" / e(a+ix)wdw+_ -oo J^ 1 27r(a + ix) a 2 7r(a + a. t)->S .4).4) Example 7.Solution of the Black-Scholes Equation 119 should depend on the variables and constants in the original equation and not on the frequency variable w.3 Changing Variables in the Black-Scholes P D E To recap the partial differential equation and its side conditions for a European style call option. F{S. Some people prefer each of the forward and inverse transforms to be scaled by a factor involving ^/2TT as in the following. T) = (K. (7.

T h e first condition implies t h a t if the stock itself becomes worthless before maturity. x. (7. namely Ce > S — Ke~rT.13) in the Black-Scholes Eq. Otherwise the call option is worth the excess of the stock's value over t h e strike price.7).8) (7. T h e second part of this condition follows from the European call option property discussed in Chapter 6.11) T h e reader will be asked t o verify t h a t Ft Fss Ka2 „-2z and \Vxx (7. (7. Substituting the results in Eqs.14) . S = Kex « • x = In ^ (7. namely the H e a t E q u a t i o n . the call option is also worthless. in only a few steps we can convert Eq.10)-(7. In fact.5) and simplifying produces t h e equation vT = vxx + (kwhere k = 2r/a2 (see exercise 6). An investor could buy t h e stock for nothing and then let the call option expire unused. l)vx .5).5) into a more well known partial differential equation. Thus as S — oo the value of the call option becomes S asymptotically.T))S = K (VXXS + vtts) = K (vx^ + o\ = e~xvx (7.t) = KV{X. > A judicious change of variables for the Black-Scholes equation can simplify it. Fs = (Kv(x. and t are defined in terms of the new variables v.kv (7.10) t = T-^^r=^-(T-t) F(S.T) How are t h e Black-Scholes equation and its side conditions altered by this change of variables? The multivariable form of the chain rule can be used to determine new expressions for t h e derivatives present in Eq.12) (7-13) K '*) in exercises 4 and 5. Suppose F. and r as in the following equations. (7. T h e b o u n d a r y conditions are specified in Eq. (7.9) (7.6) states t h a t when the expiry date for t h e option arrives. the call option will be worth nothing if t h e value of the stock is less t h a n t h e strike price. (7.120 An Undergraduate Introduction to Financial Mathematics T h e final condition given in (7. S.

t) S—^oo = lim KV(X.T) x—^oo = Kex =^ lim v{x.19) — 0 as x —> —oo and > x v(x. r G (0.T)+2aux+uxx) ([3u(x.1)+ V(X.21) (7. ^ ) The independent variable x can be thought of as corresponding to a spatial variable.14).Solution of the Black-Scholes Equation 121 The final condition for F is converted by this change of variables into an initial condition since when t = T.17) (7-18) (7.1)+ Since lims^ 0 + 0= x = — °°: then = lim KV(X.t) g_>0+ Likewise since as linig^oo x — oo.T) for x G (-oo.r)^e as x ^ o o . lim F(S. 0) = (ex .0) is then found to be Kv(x. S = lim F(S. ^|-)(7. Now another change of variables is needed. x—>oo Thus we have derived a pair of boundary conditions for the partial differential equation in (7.T)+ux) (a u{x.16) produces uT = (a2 + (k . r = 0. V{X. now x can be any real number.T)=0.15) = K(ex .20) (7.1)+ v(x.23) vx=e Vxx (au(x.0) = =F(S.k . The initial condition v(x. (7.l)ux + uxx (7.r) = ex.T) (S-K)+ (7. T) + uT) 2 ax+pt & ax+f3t vT = e Substituting the expressions found in (7. 0) = (e x .16) v{x. If a and (3 are constants then we can introduce the new dependent variable u.j3)u + (2a + k .T) = eax+l3tu(x.22) (7.24) . Thus the original Black-Scholes initial.T) x—> — oo =$• lim x—> — oo V(X. boundary value problem has been recast in the form of the following.l)a . The reader should note that where previously the price of the security S was assumed to take on only non-negative values. oo).23) into Eq. vT = vxx + (k — l)vx — kv for x G (—oo. oo) (7.20)-(7.T) ax+l3t (7. r 6 (0.

122

An Undergraduate Introduction

to Financial

Mathematics

Since a and /3 are arbitrary constants they can now be chosen appropriately to simplify Eq. (7.24). Ideally the coefficients of ux and u would be zero. Solving the two equations: 0 = a2 + (k - l ) a - k - /3 0 = 2a + k-l

yields a = (1 — k)/2 and (3 = —{k + l ) 2 / 4 . The initial condition for u can be derived from the initial condition given in (7.17). v(x,0) = ( e x - l ) + eaxu(x,0) = u{x,0) = e C=-iW2( e * _ i)+

=(fc-i)z/2

/ ex - 1 if x > 0, 0 ifx<0. - e ^ - 1 ) - / 2 if x > 0, if x < 0.

6

(fc-i)x/2 f exe^k'^xl2 |0

p(fe-iW2fe( fc+1

=

\ 0

W2-e(fc-1W2ifa;>0, if x < 0.

+

= ( e (fc+l)x/2 _ e(k-\)x/2-s

Likewise we can derive boundary conditions at x = ±oo for u from Eqs. (7.18) and (7.19). To summarize then the original Black-Scholes partial differential equation, initial, and boundary conditions have been converted to the following system of equations. uT = uxx u(x, 0) = (e

{k+l

**for x G (—00,00) and r S (0,T<r 2 /2) >'
**

2

(7.25) (7.26) (7.27)

- e (fc-iW )+

2

for

x G 2

(-oo, 00)

u(x, r) -^ 0 as x -> ±00 for r G (0, T<r /2)

The PDE of Eq. (7.25) is the well-known heat equation of mathematical physics. The Fourier Transform technique introduced earlier will now be used to solve this initial boundary value problem.

7.4

Solving the Black-Scholes Equation

We begin by taking the Fourier transform of both sides of the heat equation (since the Black-Scholes equation has been through two changes of variables

Solution of the Black-Scholes

Equation

123

**and has been converted into the heat equation on the real number line) in Eq. (7.25).
**

T{ur)

OO

=

TT{UXX\

rOO

/

-oo

uTe~iwxdx u{x,r)e~lwx

= \

J — oo

uxxe~iwxdx u{x,r)e

J ~ oo

—

®^

J — oo

dx = (iw)2

2

— = -w u (7.28) dr The reader should note that in deriving Eq. (7.28) Theorem 7.1 is used. This last equation is an ordinary differential equation of the type used to model exponential decay. Separating variables (see Sec. 5.3, especially Eq. (5.17) and the paragraphs which follows it) and solving this equation produces a solution of the form

U(W,T) =De~w2T.

The expression represented by D is any quantity which is constant with respect to r. Even though this equation was solved using ordinary differential equations techniques, the quantity u is a function of both w and r. To evaluate D we can set r = 0 and determine that u(w, 0) = D. Thus D is merely the Fourier transform of the initial condition in Eq. (7.26). For simplicity of notation we will write D = f(w). Thus the Fourier transformed solution to the heat equation is u(w,T) = f(w)e-w2T. (7.29)

Now this solution must be inverse Fourier transformed and then have its variables changed back to the original variables of the Black-Scholes equation.

U(X,T)

= ( e ( fe +!W 2 - e ( f c - 1 )^ 2 )+ *

1

2v f°° (e(*+Di _ e ( * - D i ) + e - ^

7rr

-J—e-*2/^

dz (7.30)

2JTXT

The reader should note that the Fourier convolution theorem was used and also the result of exercise 3. If the substitution z = x + \Z2ry is made then

124

An Undergraduate Introduction

to Financial

Mathematics

Eq. (7.30) becomes

U(X,T)

=

_L_ /

Lik+Vix+V^y)^

_ e(k-l)(x+V2^y)/2y

e-y

2

/2

dy,731^

e

e

27T Jk-l)x/2Jk-l)2T/4 V27T

f

J-x/\ roo

e-(y-H^)v^r/2dy

(7-32)

J-x/V2^

T h e reader will be asked to fill in the details of this derivation in exercises 7 and 8. T h e two improper integrals of Eq. (7.32) can be expressed in terms of the cumulative distribution function for a normal random variable. We will make the substitution w = y — |(fc + 1 ) \ / 2 T in the first integral and w' = y — |(fc — l ) \ / 2 r in the second. T h e first integral of Eq. (7.32) equals the following expression

fx/VZF+Hk+VV^

/

J-oo

e~w/2 dw = V2^(f> [—= + -{k +

VV2T 2

/

T

1

,

'

l)\^

where 4> is the cumulative distribution function for a normal r a n d o m variable with mean zero and s t a n d a r d deviation one. T h e reader should verify this calculation (exercise 9) and a similar calculation for the second improper integral of Eq. (7.32). T h u s far the solution to the initial b o u n d a r y value problem in Eqs. (7.25)-(7.27) is given by

u(x, r ) = e (>=+iW2+(M-i) 2 T/4 0 (* + hk V V 2r 2

s(/c-lW2+(/c-l)V40

+ 1 ) V

^A J .

(733)

(*

+

1 {k _

1)y/£f\

Now we begin the task of reconverting variables to those of the BlackScholes initial b o u n d a r y value problem as stated in Eqs. (7.5)-(7.7). Using the change of variables in Eq. (7.20) this solution can be re-written in terms

Solution of the Black-Scholes

Equation

125

of the function

V(X,T).

V(X,T)

= e-(i-l)x/2-(fc+l)'x/4t((liT)

= ^(vi + ^ + 1 ) V 5 7 )

-"**(^ + 5(*-1),/5f

(7.34)

Now using the change of variables described in Eqs. (7.8) and (7.9) the reader can show in exercise 11 that

^ J L . l l U D V g . ^ ' ^ ^ - "

w/

x

(7.35)

(7.36)

V^

+ l{k2

1)V2T = W- cry/T - t.

**Using these in Eq. (7.34) yields
**

V(X,T)

= ^4>{w) - e _ r ( T - ' V ( ' « "

<ry/T^t)

which upon using Eq. (7.10) produces the sought after Black-Scholes European call option pricing formula. For the sake of simplicity and meaningful notation the value of the European call will be denoted C. C{S, t) = S<j>{w) - Ke-r(T-t](t>{w - aVT^t) (7.37)

A surface plot of the value of the European call option as a function of S and t is shown in Fig. 7.1. The value of a European put option could be found via a similar set of calculations or the already determined value of the European call could be used along with the Put-Call Parity formula (6.1) to derive P(S, t) = Ke-^-^^oy/T - t - w) - S<t>{-w). (7.38)

A surface plot of the value of the European put option as a function of S and t is shown in Fig. 7.2. Example 7.3 Suppose the current price of a security is $62 per share. The continuously compounded interest rate is 10% per year. The volatility of the price of the security is a = 20% per year. The cost of a five-month European call option with a strike price of $60 per share can be found using Eqs. (7.35) and (7.37). If we summarize the quantities we know, in

126

An Undergraduate Introduction

to Financial

Mathematics

S(0)

Fig. 7.1

A surface plot of Ce as a function of S and t.

the notation of this section, we have: T = 5/12, £ = 0, r = 0.10, a = 0.20, 5 = 62, and K = 60.

Thus we have w « 0.641287 and therefore the price of the European call option is C = $5,789. Example 7.4 Suppose the current price of a security is $97 per share. The continuously compounded interest rate is 8% per year. The volatility of the price of the security is a = 45% per year. The cost of a three-month European put option with a strike price of $95 per share can be found using Eqs. (7.35), and (7.38). If we summarize the quantities we know, in the notation of this section, we have: T=l/4, t = 0, r = 0.08, a = 0.45, S = 97, and K = 95.

Thus we have w as 0.293985 and consequently the put option price is P = $6.71343.

2 A surface plot of Pe as a function of S and t. 7.5 Exercises (1) Let the function f(x) be defined as 1 " e~ax if x > 0 0 otherwise where a is a positive constant. Show that the inverse Fourier Transform of f is f(x) = 0 . . Show that the Fourier Transform of / (2) Let the function f(w) be defined as /(to) = < „ if x f e~aw if w > 0 [0 otherwise where a is a positive constant. (7. (7.8).10). 7. (7.12).9). 1 • ^. verify the identity shown in Eq. J J \ I 27r(a—ix) (3) If o is a positive constant verify that the inverse Fourier Transform of 2 e-<™ is -4= e .Solution of the Black-Scholes Equation 127 S(0) Fig.x 2 / 4 <\ (4) Using the new variables described in Eqs. (7. and the multivariable chain rule.

128 An Undergraduate Introduction to Financial Mathematics (5) Using the new variables described in Eqs. (7. (8) Use completion of the square in the exponents of the integrand in Eq.10) and the partial derivatives found in (7. and the multivariable chain rule. (7. and (7. verify the identity shown in Eq. (9) Using the change of variable w = — y + \(k + l ) \ / 2 r in the first improper integral of Eq.8). (7. and (7.13).32) verify that 1 1 r°° '2n ' J~X/V2 e-(y-i(k+l)V^) 2 f2dy dw e 2 (x/V^+Uk + l)^ where (j> is the cumulative distribution function for a normal random variable with mean zero and standard deviation one.32).10).14) by using the variables described in (7. (7. (7. .11).9).13). (7. the expression for Fs shown in (7. (7.8). (7.11). (10) Using the change of variable w' = — y + \{k — l)v / 2r in the second improper integral of Eq. (7) Verify that the expression /e(fc+i)(x+.32) verify that '"" -(y-i(fe-l)v^)2/2 e-(y-i(K-i-)V2T)-/-z >2-K J-x/y/2TO'2 1 dy e 2" dw' (x/V2r+^(k- 1)V2T where <f> is the cumulative distribution function for a normal random variable with mean zero and standard deviation one.9). (7. (6) Verify that the Black-Scholes partial differential Eq. (7.31) to derive Eq. (7. (7.12)./5Fy)/2 _ Jk-l)(x+V2^y)/2~y2~y /2^ (K-i)(x-t-v<tTy)/z e e I*^J2T —N + is non-zero whenever y > —X/\/2T.5) can be simplified to the form shown in Eq.

(7. . f(S. and the exercise time is three months.Solution of the Black-Scholes Equation 129 (11) Verify using the change of variable formulas in Eqs. and the exercise time is six months? (14) A European call option on a stock has a market value of $2.9) and the fact that k = 2r/a2 that X/^?+ 1 -(k 2 + l)^ = HS/K)HrJ^/2)(T-t) MS/K) + (r+f/2)(T-t) <j\JT — t X/^P+ 1 -(k-l)^= _aV¥-t (12) What is the price of a European call option on a stock when the stock price is $52. (7.t) = S solves the Black-Scholes equation.5) that the price of the security itself.8) and (7. The stock price is $15. the strike price is $50. and the exercise time is three months? (13) What is the price of a European put option on a stock when the stock price is $69. What is the volatility of the stock? (15) Show by differentiation and substitution in Eq. i. the interest rate is 5%. the interest rate is 5%. the interest rate is 12%. the strike price is $13. the strike price is $70.50.e. the stock's volatility is 30%. the stock's volatility is 35%.

This page is intentionally left blank .

Since the valuation of European-style options depends heavily on the cumulative distribution function of the standard normal random variable and a composite variable w. To mathematicians the word derivative means an instantaneous rate of change in a quantity. The derivative with respect to t of quantity w 131 . To a quantitative analyst a derivative is a financial entity whose value is derived from the value of some underlying asset. 8.1 Theta The quantity 0 is defined to be the rate of change of the value of an option with respect to t. we will begin by investigating their partial derivatives. Members of the quantitative financial profession refer to the subject matter of this chapter as the "Greeks" since a Greek letter is used to name each derivative (except for one which we will meet in due time). Hence a European call option is a derivative who value is a function of (among other things) the value of the security underlying the option. In this chapter we explore derivatives from the mathematician's perspective. The material of the present chapter will be use extensively in Chapter 9. In the Chapter 9 we will use these derivatives in the manner of a quantitative analyst. We will calculate 8 for the European call and put options derived in Chapter 7. Knowledge of these sensitivities allows a portfolio manager to minimize changes in the value of a portfolio when underlying variables such as the risk-free interest rate change.Chapter 8 Derivatives of Black-Scholes Option Prices Now that the solution to the Black-Scholes equation is known we can investigate the sensitivity of the solution to changes in the independent variables. An understanding of the sensitivity of an option's value to changes in its independent variables will also provide a new way of interpreting the Black-Scholes equation itself.

(7.aVT ..' V ( w . K = 245.025.37)) is = S<f>'(w)^ . Chap. \f2-K (8.t) dw _dt ~ (r<Kw 2^/T^t dw (ry/T=tj\ ~dt + ° f\n{S/K) \ T-t ° </>'\w _ T a 2 dt ay/T^i) = (s</>'(w) " Ke-'V-'Wiw .5%.K e .132 An Undergraduate Introduction to Financial Mathematics defined in Eq. consider a European call option on a stock whose current price is $250.2) Hence using the Chain Rule [Stewart (1999). 4] the cumulative distribution function for the standard normal distribution has derivative tf{x) = ^Le"* 2 / 2 .2 / 2 _ Ke-r(T-t)-(W-*VT=tf/2 . a = 0. 2a^/2Tr(T -1) ( .a2. Chap.r . r = 0.VT^)+ VMr_() j (8.3) produces 6 = — = -19.aVT^t) 2aVT^t .ay/Y^t) _ S<//(w) .9836.Ke-r(T-t) 0C = ay/T=t)\ ' (r<j){w .3) To take an example.ay/T .Ke~r(-T~V (rcj>{w .tU 5 e . and t = 0 into Eq.av'T . Thus substituting S = 250. and the strike time is four months.35) is given by dw 1 (HS/K) dt -2aVT=t{ T-t T /2 ° \ J- (8 -1} By the Fundamental Theorem of Calculus [Smith and Minton (2002). the annual risk-free interest rate is 2..Ke^-V + (j)1(w . The strike price is $245. T = 1/3. (MSIK) 2 . 2 \ T-t ' /2\ np-(w-ajT=if -^-'>{r^-. (8.r < r . (7. the volatility of the stock price is 20%.20.„ —^-i-—.t) + $ {w . 3] for derivatives we can see that the time rate of change in the value of a European call option (defined in Eq. at .

The passage of time t.4) /i f\n{S/K) T-t . The astute reader may recall that A was introduced in Sec. (8.5) will be put to use in the following example.27.Derivatives of Black-Scholes Option Prices 133 The case that 0 is negative for an option is typical since options lose value as the expiry date approaches.r -w) /: 2a V '27r(T . T = 1/4. A. and t = 0 into Eq. 8.5) Equation (8. the volatility of the stock price is 27%. Thus substituting S = 325. is unique among the derivatives covered in this chapter. the annual risk-free interest rate is 3.(w) + (V(W) .7484. This rate of change is called Delta. Delta was used to eliminate a stochastic term from Eq.Ke-^T^4>'(W . | | = </.aVT^t)) g (8. The clever choice of A left us with a deterministic partial differential equation to solve rather than a stochastic differential equation.6) . is the only non-stochastic variable among the list of independent variables upon which C or P depend. The reader will be asked to provide the details of the following derivation.2 Delta The value of an option is also sensitive to changes in the price of the underlying stock.( ' T-t r + a2/2) (8.t ) \ + Kre -^'-^{ay/T-t- -72) 2 Ke~ •r(T-t)-(w~aVT^t) 2a^2n(T -t) n(S/K) . r = 0. dP &P = ^ Se-w 2 (8.37). K = 330. a = 0.3 during the derivation of the Black-Scholes partial differential equation. 6. (6. Consider a European put option on a stock whose current price is $325. dt The derivative 0 .035. (7.5) produces dP e = — = -28. and the strike time is three months. The calculation of 0 for a European put option is carried out in a similar fashion to that for the call option.5%.12). The starting point for deriving A is again Eq. The strike price is $330.

EeHs/K)\ S J er(T-t)+ln(S/Ar) + ay/27r(T .. (8.1)) provides a convenient shortcut.2) and the following result.7) into Eq. and the strike time is 6 months. Under these conditions w = —0.2) and (8. the strike price is $110. 3P_ ds Ap = + _ ac ~ ds (8 9) dP ~dS= ^w) " * ' . (6. (8. Suppose the price of a security is $100. dw 1 dS dC .) + = 0(u)) f f^ 2 2 e-w /2 (1 _^ £ _ (i .( -*)c_(c72(T_t)_2(r+(T2/2)(r_0+21n(Jf/ig))/2\ r T = fty. the annual volatility of the stock price is 23%. Delta for a European put option could be calculated directly from the definition for P. (Se~w2'2 aSVT^t ! J. the Put-Call parity formula (Eq. the risk-free interest rate is 4% per annum.134 An Undergraduate Introduction to Financial Mathematics We can make use of Eq.8) We can put this formula to use in the following example.351325.t) dS v ^(w) + \^ V^TT J aSVT^t p-(^(T-()-2™VT=t)/2 (w) + " ay/2n(T M + av/2n(T 1 e-^2/2 .t ) \ = </>{w) (8. (8.t) _ JCe.381747 and dC A c = — = 0.7) Substituting Eqs.6) produces s xe-(^)e-(»--VT=t)2/2\ V2TT I 1 _ 11^ . however.

(8. Suppose we consider a European call option on a security whose value is $295.9) that d2P _d2C e~w^2 ~ dS2 ~ dS2 . and its T as follows. As the price of the security increases. the volatility of the security is 25% per annum. (8. its A. the annual risk-free interest rate is 4.Derivatives of Black-Scholes Option Prices 135 Thus using the same parameter values as in the previous example. C = 15. Co So as the European call option increases in value with an increase in the value of the underlying security. Thus for a European call option T is d2C dS2 d fdC dS \ dS d Mw)) ds (using Eq. then T measures the concavity of the value of the option as a function of S.7173. the price of the underlying security.351325 .3 Gamma The g a m m a T. 8.1 ) We can readily see from Eq. We can calculate the price of the option. of an option or collection of options is defined to be the second derivative of the option with respect to S. Hence T is the partial derivative of A with respect to S.10) crSy27r(T . T = 0. the . § ^ = 0.1 = -0.aSy/2n(T-t)' (8 "U) If all independent variables are held fixed except for S.613297.0127122 Thus the value of the option will increase with the price of the underlying security at an increasing rate. A c = 0. and the expiry date is two months.8)) w2/2 / j /2~7r p-w>/2 \aSVT-t (using exercise 4) (8. the put option decreases in value.25%. The strike price is $290.648675.

(7. 8. For a European call option defined as in Eq. This will have important implications for the practice known as hedging which will be explored in Chapter 9. C 191 17 16 15 14 292 294 296 298 300 Fig. dw ~do~ sjT-t . Vega comes to us through medieval Latin from Arabic. In terms of elementary calculus.12) where.w/a.4 Vega The vega of an option is the change in the value of the option as a function of the volatility of the underlying security.37).1. 8. 8.13) . Among the partial derivatives of the value of an option vega distinguishes itself as the only partial derivative given a non-Greek letter name. the graph of C as a function of S (holding all other variables fixed) is increasing and concave upward on an interval containing S = 295. (8.Ke-'W-Qfiiw da aVT~^t) f dw \ ~da~ VT-t .136 An Undergraduate Introduction to Financial Mathematics price of the option will increase at an increasing rate. dC_ .1 Since the function C(S) is increasing and concave upward on an interval containing S = 295 the price of the European call option increases at an increasing rate when the price of the underlying security is near $295. dw S(j)'{w)— . See Fig. according to exercise 8.

13) into Eq.^^e- 2 / 2 fs _ KeHs/K)\ sVr^tcp'(w) ~ -^—e-w2i2{s .t(j>'(w) w e-w2l2 ay/^n (g __ W Ke-r(T-t)+waVT=i-a (T-t)/2 = SVT=i<l>'(w) X (S . It is easily seen from the Put-Call Parity Formula (Eq.s) a\/2TT SVT-t /2TT w !1 e~_w2/2 (8. (8.ff) SVT~t<P'(w) .(s</>'{w) .S/. and whose expiry . (8.V2_ V27T (816) Thus the vega for a European style option (either call or put) on a security who current value is $160. (6.15) in other words the rate of change in the value of a European put option with respect to volatility is the same as the rate of change in the value of the corresponding European call option with respect to volatility.Derivatives of Black-Scholes Option Prices 137 Substituting Eq.. Thus we may unambiguously define vega for a European style option to be v=5VTEte_„.14) Note that the value of the European call option increases with increasing volatility in the underlying security.Ke-^-^fiiw - ay/T^t)^ ' Ke-r(T~t)e-(w-aVT=if/2\ Sy/T . whose strike price is $150.1)) that dP da Sy/T=t V27T _w2/2 e-w l\ (8.12) produces dC s<f>'(w) (yr=iSy/T=i<l>'(w) .~^= ay/2TT ^) + Ke-^^H'iw -aVr~i) (Se~w2/2 \ 2 (^) Sy/T=i(l>'(w) .^ -^—e-w'2 + (r+CT2/2)(T-t)-<x2(T-t)/2\ e -r(T-i)+ln(.

r ( T .138 An Undergraduate Introduction to Financial Mathematics date is 5 months can be calculated using Eq.aVT^t)) ^ + (8.16).t)e. (7.= (S<p'(w) .Ke-^-^'iw .£)e.35) we see that dw _ y/T^t dr a Substituting Eq.r ( T .5 R h o The rho of an option is the rate of change in the value of the option as a function of the risk-free interest rate r.i ) e . Once again the European call option will be used to demonstrate the calculation of rho. Suppose the volatility of the security is 20% per year and the risk-free interest rate is 2.t ) e .t ) </»(^ .37) it is seen that ^ = ( W H .Kel^s/KA V „-W2/2/rp ) C W 2 aV2n ~ ' ^r~ t + K(T .^ or V / a r(T + K{T .75% per year. 8. (7.r(T .Ke~r{T~t)+waVT^i'{T''t)a2/2) V / + K(T .vy/T=ij) ^ .Ke-^T-^4>'{w .653219 and V = 33.2866.t) ^ aV2n (S .<ry/T=t) (S .oy/T^t) (8.t)e-^T-V<t>{w *y/T=i) aVT^t) / _ J.r ( T .Ke-HT-t)+HS/K) + (r+*2/V(T-t)-(T-t)a2/2\ e ~W *\/T-t \ + K(T .i)e.aVT .17) yields d (8. Using Eq.* V ( w .r ( T -*V(w . Then w = 0.t ) 0 ( w - aVT^t) W /2 T e (s . (8.i)e.-*V(™ aVT^t) (Se-w2'2 V Ke-r(T-t)e-{w-*VT=lY/2\ / x/T^t y_ ay/2-K l + K(T .aVT^t) Calling on the definition of w given in Eq.17) K(T . (8.18) -§.t)e-^T-^4>{w pc = KiT .18) into Eq. ay/lit (S _ s)- ^ O\J2-K ^ + K(T .t} 4>(w .19) . (8.

20) In the last equation above we made use of the fact that <f>(x) = 1 — 4>(—x) for the cumulative distribution function of a normally distributed random variable. Under these conditions. or 8. w = 0. This equation is repeated below for convenience. To put Eq. namely e = Ft.14).w) (8. the expiry date is three months.1)) and Eq.t ) e . (8.= -20.K{T aVT^t)^j t)e~r(T~^ V ( < r \ / T . while the strike time is 4 months. In this chapter we have given names to the partial derivatives Ft. consider the situation of a European put option on a security whose underlying value is $150.25%. Fs. The strike price is $152.0064434 and ^ = 33. the volatility is 15% per annum. and FSs. 0 .t)e-<T~^4>{w . (8. The strike price is $230.<t>(w .6 Relationships Between A. The volatility of the price of the security is 27% annually and the risk-free interest rate is 3. = K(T .20) to work in an example.t .' ) ( l .Derivatives of Black-Scholes Option Prices 139 Consider the situation of a European call option on a security whose current value is $225.4156. Then w = —0.0557697 and dP -—.5%. rF = Ft + rSFs + -<J2S2FSS This equation assumes the risk-free interest rate is r and the volatility of the security is a.r ( r . and T The value F of an option on a security S must satisfy the Black-Scholes partial differential Eq. and the risk-free interest rate is 2. (6. (6. or The rho for a European put option can be determined from the Put-Call Parity Formula (Eq. A = FS. ss- .19).aVT^t) = -K(T pP = -K{T .i)er(T-t .8461.

10) Eq. T h e m a t h ematical aspect of hedging is made possible by the linearity property of the Black-Scholes partial differential equation.16) Eq.21). (8.1 Eq. with then should be of large magnitude then T must be large and of opposite Returning to Eq. W h e n the value of the option is insensitive to the passage of time 0 = 0 and Eq. 0 . (8. (8.1 Listing of the derivatives of European option values and some of their properties. In the next chapter the practice of hedging will be described. This table is provided as a convenient place to find the definitions of these quantities in the future. (8. (8. calculation of 0 can be used to approximate F.19) Eq. and F. a collection which may include stocks. In practice. put and call options. etc. Name Theta Delta Gamma Vega Rho Symbol Definition Eq. T h u s 0 a n d T t e n d t o be of opposite algebraic sign. (8. Y in the domain of L.3) Eq.140 An Undergraduate Introduction to Financial Mathematics Table 8. Table 8. if the rate of change in the value of the option respect to the value underlying security is zero (or at least very small) the Black-Scholes equation becomes rF = e+^a2S2T. An operator L[-] is linear if L[aX + Y] = aL[X] + L[Y] for all scalars a and vectors X.5) Property ec eP V Ac AP <p(w) 4>(w) .1 summarizes the option value derivatives discussed in this chapter and lists some of their relevant properties. In the case of the Black-Scholes P D E the vectors are solution . (8. Hedging is performed on a p o r t f o l i o of securities. (8. (8.21) becomes If A sign. cash in savings accounts.20) Ac > 0 AP <0 r r>o V>0 pc > 0 pp < 0 PC PP T h u s the Black-Scholes partial differential equation can be re-written as rF = 0 + rSA + )-a2S2T.21) T h e value of the option is a linear combination of A.

(9) Calculate the V of a European style option on a security who current value is $300. (8. The strike price is $140 while the expiry date is eight months. (7. whose strike price is $305.5%.Derivatives of Black-Scholes Option Prices 141 functions. (8. The volatility of the stock is 15% annually. (2) Carefully work through the details of the derivation of 0 for a European put option. The strike price is $175.35). . the volatility in the price of the security is 30% annually.75% per year. The annual risk-free interest rate is 5. The strike price is $165 while the strike time is five months. The annual risk-free interest rate is 3%. (3) Find 0 for a European put option on a stock whose current value is $275. to the Black-Scholes partial differential Eq. The volatility of the stock is 20% annually. The volatility of the stock is 22% annually. The annual risk-free interest rate is 2%. (6. (4) Find the partial derivative of w (as defined in Eq. If we define the operator L[-] to be L rrxrl dX adX ^ = iH M r ^-rX> +rS 1 2a2d 2 + s X then any solution X. the annual risk-free interest rate is 3. 7.13) from Eq.5%. (6) Find dP/dS for a European put option on a stock whose current value is $125. 8. (7) Calculate T for a European put option for a security whose current price is $180. The strike price is $310 while the strike time is three months. The strike price is $265 while the expiry date is four months. Confirm that you obtain the result in Eq.5). The volatility of the stock is 25% annually. Suppose the volatility of the security is 25% per year and the risk-free interest rate is 4. and whose expiry date is 6 months.75%. (5) Find A for a European call option on a stock whose current value is $150. and the strike time is four months. The annual risk-free interest rate is 2.14) will satisfy L[X] = 0. (8) Using partial differentiation derive Eq.7 Exercises (1) Find 0 for a European call option on a stock whose current value is $300.35) with respect toS.

whose strike price is $325. (11) Calculate the rho of a European style call option on a security who current value is $305. . and whose expiry date is 4 months. and whose strike time is 2 months.75% per year.55% per year. whose strike price is $272. Suppose the volatility of the security is 15% per year and the risk-free interest rate is 3. Suppose the volatility of the security is 35% per year and the risk-free interest rate is 2.142 An Undergraduate Introduction to Financial Mathematics (10) Calculate the rho of a European style put option on a security who current value is $270.

if the market price of the stock exceeds the strike 143 . The bank is in a favorable position if. In this chapter they will be used to hedge portfolios of investments. In fact this same quantity A.Chapter 9 Hedging Hedging is the practice of making a portfolio of investments less sensitive to changes in market variables such as the prices of securities and interest rates. an investor).14). at a predetermined time (the expiry date) in the future. If F(S. let us pause and recall what happens when a one entity (say. a bank) sells a call option on a stock to another entity (say. The stock must be available to the investor at the strike price even if the market value of the stock exceeds the strike price. the rate of return from a portfolio consisting of ownership of the underlying security and sale of the option (or the opposite positions) should be the same as the rate of return from the risk-free interest rate on the same net amount of cash. was used to derive the Black-Scholes PDE. The bank has promised the investor that they will be able to purchase the stock at a predetermined price (the strike price). In this case the call option will not be exercised and the bank keeps the money it received when it sold the option. the quantity A = Fs is central to hedging a portfolio. In Chapter 8 the partial derivatives of the values of European put and call options were calculated. The Black-Scholes equation can be thought of as the statement that when A is zero. at the strike time. (6. 9. the market price of the stock is below the strike price set when the call option was sold to the investor.1 General Principles Before launching into a discussion of hedging. Conversely. t) is a solution to the Black-Scholes partial differential Eq.

the value of this European call option is $1. As long as the stock price remains below the strike price the bank keeps its revenue from the sale of the options. The net revenue is zero when S is approximately $53. In this scenario the net revenue generated by this sequence of transactions is 1919650 + min{0. and the volatility of the stock price is 22.650. Consider this strategy: the bank creates a European call option on a stock whose current price is $50 while the strike price is $52. However the bank's losses could be dramatic when the price of the stock rises before the expiry date.5M). the bank must ensure the investor can find a seller of the stock who will agree to accept the strike price. the expiry date is four months. So that we deal with whole numbers. and if the stock price is higher the bank profits (at S = 52 the profit nearly $3.4M).9357. or they could purchase one million shares at the time the investor bought the million call options.52 . Neither of these schemes are practical for hedging portfolios in the fi- . The reader can check that when S « 48. Since the strike price is $52.144 An Undergraduate Introduction to Financial Mathematics price of the option. Notice that the present value of the stock price must be incorporated rather than just S.0M.5% per annum. considerably more than the revenue generated by the sale of the call options. the riskfree interest rate is 2.4827 the net expenditure is zero.2 illustrates the profit or loss to the seller of the European call option. Another option for the bank is to purchase the stock at the strike time and instantaneously sell it to the investor at the strike price. Figure 9. the maximum net revenue for the bank occurs at that price. S'}e. According to the Black-Scholes option pricing formula.5% per annum.50) • 106. they will expend $50M.1 shows if the stock is below that price the bank loses money (at S = 46 the loss comes to more than $2. The bank could wait until the expiry date to purchase the one million shares of the stock. If S represents the price per share of the stock at the strike time then the net revenue generated by the transaction obeys the formula 1919650+ (min{52. At a price of $56 per share the bank's net loss would be approximately $2.° 0 2 5 / 3 • 106.5 } e . One way to accomplish this would be for the bank itself to be the seller to the investor.919.0 0 2 5 / 3 . Figure 9. The bank accepts revenue equal to $1. suppose an investor purchases one million of these call options. Suppose the bank takes the latter course.91965.

In the next section a simple.10b Fig. 8. the strike price is set at $105.2 Delta Hedging The purpose of hedging is to eliminate or at least minimize the change in the value of an investor's or an institution's portfolio of investments as conditions change in the financial markets. A units of the security are bought. 9.1 Consider the case of a security whose price is $100.279806.1 The line shows the profit or loss to the seller of a European call option who adopts a covered position by purchasing the underlying security at the time they sell the option. Example 9. One variable subject to change is the value of the stock or security underlying an option. Under these conditions w = —0. and the expiry date is 3 months. The reader will recall from the previous chapter that Delta is the partial derivative of the value of an option with respect to the value of the underlying security. A portfolio consisting of securities and options is called Delta-neutral if for every option sold. nancial world due to their potentially large costs. while the risk-free interest rate is 4% per annum. practical method for hedging is explored. In Sec. the value of the option changes by A. 9.Hedging 145 . One can think of Delta in the following way: for every unit change in the value of the underlying security. the annual volatility of the stock price is 23%. the value of a European call option .2 expressions for Delta for European call and put options were derived.

the firm would receive $29615. This strategy is known as rebalancing the portfolio. the firm may choose to make periodic adjustments to the number shares of the security it holds. In this case the European call option will be exercised since the price of the security at the expiry date exceeds the strike price of $105.96155 and Delta for the option is dC A = — = 0. In that case. is C = 2.10° Fig.389813. 1 Typically an option is keyed to 100 shares of the underlying security.146 An Undergraduate Introduction to Financial Mathematics . .3. To avoid confusion we will assume a one-to-one ratio of options to shares. On the other hand. At the end of the first week the value of the security has declined to $98. The firm may choose to do nothing further until the strike time of the option. this is referred to as a "hedge and forget" scheme. ob Thus if a firm sold an investor European call options on ten thousand shares of the security 1 . 9. The example begun above can be extended to include weekly rebalancing.389813)(100) worth of the security. 9. since the price of the security is dynamic.50 and purchase $389813 = (10000)(0. most likely with borrowed money.79. so in practice options on 10000 shares would amount to 100 options.2 The line shows the profit or loss to the seller of a European call option who adopts a naked position by purchasing the underlying security at the expiry date of the option. Assume that the value of the security follows the random walk illustrated in Fig.

The horizontal line indicates the strike price of a European call on the security. Since the value of the security finishes below the strike price. Alternatively the value of the security may evolve in such as fashion that the call option is not exercised.3 A realization of a random walk taken by the value of a security.339811. Upon re-computing. In other words. 9. the firm profits from issuing the call option. A = 0.4.000. The interest for the current is added to the cumulative cost entry of the next week. the call option . Thus the investment firm would adjust its security holdings so that it now owned 3398 shares of the security with a total value of $335699.50.97. 050.50 . Notice at the expiry date the investment firm has in its portfolio 10000 shares of the security for which the investor will pay a total of $1. 9. Also at the end of the first week the firm will have incurred costs in the form of interest on the money initially borrowed to purchase shares of the security.1069460 = $10155. Table 9. Thus the net proceeds to the investment firm of selling the call option and hedging this position are 1050000 + 29615.1 summarizes the weekly rebalancing up to the expiry date. Such a scenario is depicted in Fig.1) = $299. This cost amounts to (389813)(e 004 / 52 . The reader should note that this assumes the investor does not pay for the call option until the strike time.Hedging 147 112 110 108 w 106 104 102 100 0 2 4 6 week 8 10 12 Fig.

50. In this case the firm loses a small amount of money on the collection of all these transactions. Week 0 1 2 3 4 5 6 7 8 9 10 11 12 13 S 100. Notice that at the expiry date the investment firm owns no shares of the security (none are needed since the call option will not be exercised).585915 0. If |T| is small then A is relatively insensitive to changes in S and hence rebalancing may be needed infrequently. Table 9.50 .79 102.148 An Undergraduate Introduction to Financial Mathematics Table 9.2 summarizes the weekly rebalancing activity of the investment firm as it practices Delta hedging.29669 = -$53.389813 0.52 103.86 110.40 112.46 108.41 102. The quantity discussed in Sec.67 106. In other words Gamma is the rate of change of Delta with respect to S.82 102.878690 0.347145 0.462922 0.589204 0. The net proceeds to the investment firm of selling the call option and hedging its position are 29615. Thus an investment firm can monitor Gamma in order to determine the frequency at which the firm's position should be rebalanced.17 106.1 Delta hedging using portfolio rebalancing at weekly intervals.428236 0.490192 0.05 104.595047 0.25 100.491348 0.47 A 0.08 105.460541 0. If |F| is large then A changes rapidly with relatively small changes in S. Shares Held 3898 3398 4629 4902 4605 4282 3471 5892 4913 5950 5859 8787 9858 10000 Interest Cost 300 262 359 382 358 333 271 468 390 475 468 717 811 0 Cumulative Cost 389800 340705 467169 495760 465604 432935 351625 608643 507129 617524 608366 932085 1053247 1069460 expires unused.3 known as Gamma (r) is the second partial derivative of the portfolio with respect to the value of the underlying security. . 8. Rebalancing of the portfolio can take place either more or less frequently than weekly as was done in the two previous extended examples. In this case frequent rebalancing of the investment firm's position may be necessary.985811 10 .00 98.339811 0.

3 Delta Neutral Portfolios In exercise 15 of Chapter 7 the reader was asked to verify that a stock or security itself satisfies the Black-Scholes PDE (6. Thus it reasonable to contemplate Delta for the portfolio. The quantity V satisfies the Black-Scholes equation since C and S separately do. (9. The horizontal line indicates the strike price of a European call on the security. 9.4 Another realization of a random walk taken by the value of a security. The partial derivative of the entire portfolio with respect to S represents the sensitivity of the value of the portfolio to changes in S.Hedging 149 Fig. Differentiating both sides of Eq.1) where So is the price of the security at the time the hedge is created.For . The portfolio contains a short position in a European call option and a long position in the security (hedged appropriately as described in the previous section).14). and the equation is linear. Thus the net value V of the portfolio is V=C-AS=C dC_ as So s.1) with respect to S produces dV OS dC ~dS So This quantity equals zero whenever S = SQ (for example at the moment the hedge is set up) and remains very close to zero so long as S is near So. Now suppose that a portfolio consists of a linear combination of options and shares of the underlying security. 9. (9.

001508 Shares Held 3898 4406 3868 3946 4827 2462 Interest Cost 712 430 504 786 1102 0. T h e t e r m involving 9 is not stochastic and thus must be retained. dVln 0 N d2V (S - S0)2 v = r0 + ^(t-to) + -(s-s0) + ^—J-L + - sv = est + ASS + ^r(ss)2 + • • • All omitted terms in the Taylor series involve powers of St greater t h a n 1. .078574 0.036209 0.91 103. (9. however.95 91.37 97.482725 0.40 95.00 101.81 95.71 100.071229 0.45 97.440643 0. (5.58 95.150 An Undergraduate Introduction to Financial Mathematics Table 9. if the makeup of the portfolio can be adjusted so t h a t r = 0. Assuming t h a t the risk-free interest rate remains constant and the volatility of the security does not change then a Taylor series expansion of the value of the portfolio in terms of t and S yields ^ ^ dV. If the portfolio is hedged using Delta hedging then A for the portfolio is zero and thus sv^est + ^r(ss)2.e.12 92.38). ek 0 1 2 3 4 5 6 7 8 9 10 11 12 13 S 100.246176 0.394649 0.0 362 15 0 300 340 299 305 375 198 74 54 60 80 104 49 24 0 Cumulative Cost 389800 441769 388077 396247 487621 256959 96244 70623 77545 104521 135491 64126 31072 29669 this reason a portfolio which is hedged using Delta hedging is sometimes called D e l t a n e u t r a l .110154 0.43 100.043022 0.75 96.69 91.10 A 0.389813 0. i. the approximation can be further refined if the portfolio can be made G a m m a n e u t r a l .2 Delta hedging using portfolio rebalancing at weekly intervals for an option which will expire unused.386757 0. see Eq.2) where the approximation omits terms involving powers of St greater t h a n 1.050427 0. . T h e G a m m a t e r m is retained since the stochastic r a n d o m variable S follows a stochastic process dependent on vSt.

03618. (8. the portfolio can be made Delta neutral with the appropriate combination of the option and the underlying security. An investment firm sells a European call option on this stock with a strike time of three months and a strike price of $102. Let the number of the early option sold be we and the number of the later option be wi.3) Example 9. An investment firm may sell a number of European call options with the expiry date three months hence and buy some other number of the European call options on the same stock with an expiry date arriving in six months. However. where Te and Ti denote the Gammas of the earlier and later options respectively. The reader should keep in mind that the underlying security will have a V = 0 and thus Gamma neutrality for the portfolio will be maintained while Delta neutrality is achieved. (9.22 and the risk-free interest rate is 2. The portfolio can be made Gamma neutral by manipulating a component of the portfolio which depends non-linearly on S. the underlying security can be added to the portfolio in such a way as to make the portfolio Delta neutral. The portfolio cannot be made gamma neutral using a linear combination of just an option and its underlying security. The relative weights of the earlier and later options can be chosen so that T-p = 0.02563. (9. As explained in the previous section.10) the Gamma of the three-month option is T 3 = 0. The . the portfolio cannot contain only the option and its underlying security. The firm buys European call options on the same stock with the same strike price but with a strike time of six months. One such component is the option. Thus Eq. One way to achieve the goal of a Gamma neutral portfolio is to include in the portfolio two (or perhaps more) different types of option dependent on the same underlying security. According to Eq.Hedging 151 9. while that of the six-month options is T 6 = 0. For example suppose the portfolio contains options with two different strike times written on the same stock. The Gamma of the portfolio would be T-p = weTe .wiTi. since the second derivative of S with respect to itself is zero. as mentioned earlier.5% per year.4 Gamma Neutral Portfolios Gamma for a portfolio is the second partial derivative of the portfolio with respect to S.2 Suppose a stock's current value is $100 while its volatility is a = 0.2) for a Gamma neutral portfolio reduces to SP « est. Once the portfolio is made Gamma neutral.

152 An Undergraduate Introduction to Financial Mathematics portfolio becomes Gamma neutral at any point in the first quadrant of u>3u. The present chapter has focused mainly on making a portfolio's value resistant to changes in the value of a security. Thus prior to the inclusion of the underlying stock in the portfolio. Suppose W3 = 100000 of the three-month option were sold. In the present chapter it was also . This discussion of hedging is far from complete. The quantities Rho and Vega discussed in Chapter 8 can be used to set up portfolios hedged against changes in the interest rate and volatility. then the portfolio is Gamma neutral if w6 = 141163 six-month options are purchased.w6A6 = (100000)(0. 9. Therefore the portfolio can be made Delta neutral if 25038 shares of the underlying stock are sold short.51- 0.5- .03618w3 . 80 90 ^ 100 S 110 120 130 Fig. Figure 9.5r 1.5 The aggregate value of a Gamma neutral portfolio is insensitive to a range of changes in the value of the underlying securities.02563w6 = 0 is satisfied.4728) .5 shows that over a fairly wide range of values for the underlying stock the value of the portfolio remains nearly the same. 10 6 P 2.(141163)(0.6-space where the equation 0.0.5123) = -25038. In reality the risk-free interest rate and the volatility of the stock also affect the value of a portfolio. the Delta of the portfolio is W3A3 .

Hedging

153

assumed that the necessary options and securities could be bought so as to form the desired hedge. In practice this may not always be possible. For example an investment firm may not be able to purchase sufficient quantities of a stock to make their portfolio Delta or Gamma neutral. In this case they may have to substitute a different, but related security or other financial instrument in order to set up the hedge. This strategy will be discussed in the next chapter after the prerequisite statistical concepts are introduced.

9.5

Exercises

(1) A call option closes "in the money" if the market price of the underlying security exceeds the strike price for the call option at the time the option matures. Use Eq. (7.35) to show that A — 1 as t — T~ > > for an "in the money" European call option. (2) A call option closes "out of the money" if the strike price for the call option exceeds the market price of the underlying security at the time the option matures. Use Eq. (7.35) to show that A — 0 as t — T~ > > for an "out of the money" European call option. (3) Suppose the price of a stock is $45 per share with volatility a = 0.20 per annum and the risk free interest rate is 4.5% per year. Create a table similar to Table 9.1 for European call options on 5000 shares of the stock with a strike price of $47 per share and a strike time of 15 weeks. Rebalance the portfolio weekly assuming the weekly prices of the stock are as follows. Week S Week S 0 45.00 8 47.79 1 44.58 9 48.33 2 46.55 10 48.81 3 47.23 11 51.36 4 47.62 12 52.06 5 47.28 13 51.98 6 49.60 14 54.22 7 50.07 15 54.31

(4) Repeat exercise 3 for the following scenario in which the call option finishes out of the money. Week S Week S 0 45.00 8 41.94 1 44.58 9 42.72 2 45.64 10 44.83 3 44.90 11 44.93 4 43.42 12 44.19 5 42.23 13 41.77 6 41.18 14 39.56 7 41.52 15 40.62

154

An Undergraduate Introduction

to Financial

Mathematics

(5) Repeat exercise 3 assuming this time that the financial institution has issued a European put option for the stock. (6) Repeat exercise 4 assuming this time that the financial institution has issued a European put option for the stock. (7) Set up a Gamma neutral portfolio consisting of European call options on a security whose current value is $85. Suppose the risk-free interest rate is 5.5% and the volatility of the security is 17% per year. The portfolio should contain a short position in four-month options with a strike price of $88 and a long position in six-month options with the same strike price. (8) For the portfolio described in exercise 7 add a position in the underlying security so that the portfolio is also Delta neutral. (9) Set up a Gamma neutral portfolio consisting of European call options on a security whose current value is $95. Suppose the risk-free interest rate is 4.5% and the volatility of the security is 23% per year. The portfolio should contain a short position in three-month options with a strike price of $97 and a long position in five-month options with a strike price of $98. (10) For the portfolio described in exercise 9 add a position in the underlying security so that the portfolio is also Delta neutral.

Chapter 10

Optimizing Portfolios

There are several notions of the idea of optimizing a portfolio of securities, options, bonds, cash, etc. In this chapter we will explore optimality in the sense of maximizing the rate of return, minimizing the variance in the rate of return, and minimizing risk to the investor. We will also introduce the Capital Assets Pricing Model which attempts to relate the rate of return of a specific investment to the rate of return for the entire market of investments. We will see in this chapter that the difference in the expected rates of return for a specific security and the risk-free interest rate is proportional to the difference in the expected rates of return for the market and the risk-free interest rate. In order to explain this proportionality relationship we will introduce two additional statistical measures, covariance and correlation. We will also make use of these concepts in developing additional hedging strategies.

10.1

Covariance and Correlation

The concept of covariance is related to the degree to which two random variables tend to change in the same or opposite direction relative to one another. If X and Y are the random variables then mathematically the covariance, denoted Cov(X,Y), is defined as

Cov (X, Y)=E[(X-E

[X])(Y - E [Y])}.

(10.1)

155

156

An Undergraduate Introduction to Financial Mathematics

Making use of the definition and properties of expected value one can see that Cov (X, Y)=E [XY - YE [X] - XE [Y] + E [X] E [Y]}

= E [XY] - E [Y] E [X] - E [X] E [Y] + E [X] E [Y] = E [XY] - E [X] E [Y], (10.2)

where Eq. (10.2) is generally more convenient to use t h a n the expression in the right-hand side of Eq. (10.1). E x a m p l e 1 0 . 1 Table 10.1 lists the names, heights, and weights of a sample of Playboy centerfolds [Dean et al. (2001)]. This d a t a will be used to illustrate the concept of covariance. Let X represent height and Y represent weight for each woman. T h e reader can readily calculate t h a t E [X] = 67.05 inches and the E [Y] = 117.6 pounds. T h e expected value of the pairwise product of height and weight is E [XY] = 7887.85. Thus the covariance of height and weight for this sample is Cov(X,Y) = 2.77. Note t h a t the covariance is positive, indicating t h a t , in general, as height increases so does weight.

Table 10.1 A sample of names, heights, and weights for women appearing as Playboy centerfolds. Name Allias, Henriette Anderson, Pamela Broady, Eloise Butler, Cher Clark, Julie Sloan, Tiffany Stewart, Liz Taylor, Tiffany Witter, Cherie York, Brittany Height (in.) 68.5 67 68 67 65 66 67 67 69 66 Weight (lb.) 125 105 125 123 110 120 116 115 117 120

From the definition of covariance several properties of this concept follow almost immediately. If X and Y are independent random variables then E [XY] = E [X] E [Y] by Theorem 2.5 and thus the covariance of independent random variables is zero. T h e following set of relationships can also be established.

Optimizing

Portfolios

157

Theorem 10.1 Suppose X, Y', and Z are random variables, then the following statements are true: (1) C o v ( Z , X ) = V a r ( X ) , (2) Cov (X, Y) = Cov (Y, X), (3) Cov (X + Y,Z) = Cov (X, Z) + Cov (Y, Z). The proofs of the first two statements are left for the reader as exercises. Justification for statement 3 is given below. Proof. Let X,Y, and Z be random variables, then

Cov {X + Y, Z) = E [{X + Y)Z] - E [X + Y] E [Z] = E [XZ] + E [YZ] - E [X] E [Z] - E [Y] E [Z] = E [XZ] - E [X] E [Z] + E [YZ] - E [Y] E [Z] = Cov {X, Z) + Cov (Y, Z)

•

The third statement of Theorem 10.1 can be generalized as in the following corollary. Corollary 10.1 Suppose {XL, X 2 , . . . , Xn} and {Y1; Y2,..., Ym} are sets of random variables where n,m > 1.

(

n

m

\

n

m

j=i

i=i

/

i-i j=i

Proof. We begin by demonstrating the result when m = 1. The corollary holds trivially when n = \ and follows from the third statement of Theorem 10.1 when n = 2. Suppose the claim holds when n < k where fcgN. Let {Xi, X 2 , . . . , Xk, Xk+i} be random variables. C o W f ^ Y i j = C o v f ^ X ^ y i J + Cov(X fc+1 ,Y 1 ) fc = ^

i=i

Cov (Xi ,Y1)+ Cov (X fe+ i, Yi)

fc+i

= ]TCov(Xi,Y1)

Suppose the claim has been proved for m < k where again k & N..158 An Undergraduate Introduction to Financial Mathematics Therefore by induction we may show that the result is true for any finite. positive integer values of m. X2. i=l / n ^ ^ =E k Cov n E x *> Y J Cov F + Cov E x ^ F^+i n =E E J—12=1 n k (**. then n fc+1 \ /fc+1 n Cov (E -E 0 x y =Cov k £ '-'£* I n y \ / n N / ^ = ^ j=l fc Cov \ / Y j: 5 ] X. Now we can relax the assumption on m..*+i) = EECov(x-^) Therefore Eq. • i = l .4 ) .1. Let {Yi. and n..J ) + E Cov ( x .7 = 1 n fc+1 Yet another corollary follows from Corollary 10. Xn} are random variables then Var ^ Xt = Y. (10. Corollary 10. This one generalizes statement 1 of Theorem 10.1. Yfc. Y2.3) holds all finite. • • •.Var (*0 + E E Cov (**> *>) • ( 10 . integer value of n (at least when m = 1).^+1) 2=1 n =E E Cov (*<> ^ + E i=l Cov x y ( . i=l n / \ + Cov V Yfe+1. Yfc+i} be random variables.2 If {Xi..

or zero. The correlation is more than just a simple re-scaling of the covariance. \ i=l n =E E n Cov x ( *' X J ) n ( by C o r o l l a r y i0-1) n = J2 z=l n Cov (^<. while if a < 0 then p(X. the correlation always lies in the interval [—1.1 was used to reintroduce the variance in the last line of the derivation. This quantity is known as correlation and is denoted p (X. b G R with a / 0 .E*.^ ) + E E Cov (**> x i ) n z=l j ^ z n = J ] Var {Xi) + E E Cov (**' *i) Once more the first statement of Theorem 10. Y).2 Suppose X and Y are random variables such that Y = aX + b where a. Let Y = Y^i=i %i> orem 10.Y) = .1]. Y) / n \ / n n j —1 Var K ^ v«=l / = Cov n E*i. Y) = 1. Theorem 10. negative. The correlation of two random variables is defined as The correlation of two random variables can be interpreted as a measure of the degree to which one of the variables can be written as a linear function of the other. . While the covariance of X and Y may numerically be positive.1 . according to the first statement of The- Var (Y) = Cov (Y. • Often a quantity related to covariance is used as a measure of the degree to which large values of a random variable X are associated with large values of another random variable Y.1. Once again we see that independent random variables have a correlation of zero and hence are described as uncorrelated. If a > 0 then p (X.Optimizing tnen Portfolios 159 Proof.

Lemma 10.160 An Undergraduate Introduction to Financial Mathematics Proof. Before we can bound the correlation in the interval [—1.Y)~ a V a r W VVar (X) • a 2 Var (X) \a\ • which is —1 when a < 0 and 1 when a > 0. or E [y 2 ] = oo are left as exercises. E [X2] = oo.aE [X] E [X] . If X and Y are random variables Proof. Cov (X. Suppose for the purposes of this proof that 0 < E [X 2 ] < oo and 0 < E [Y2] < oo. The cases in which E [X2] = 0. If a and b are real numbers then the following two inequalities hold: 0 < E [(aX + bY)2] = a2E [X2] + 2abE [XY] + b2E [Y2] 0 < E [(aX .1 (Schwarz Inequality) then (E [XY])2 < E [X2] E [Y2]. Therefore using the result of exercise 18 from Chapter 2 p(X.6E [X] = a (E [X2] . Y) = Cov (X.1] we will need to prove the following lemma.E [X] E [aX + b] = E [aX2 + bX)] .E [X]2) = aVar (X) The reader should make note of the use of Theorem 2. . We start by calculating the covariance of X and Y.2abE [XY] + b2E [Y2] If we let a2 = E [Y2] and b2 = E [X2] then the first inequality above yields 0 < E [Y2] E [X 2 ] + 2y/E[Y2]E[X2]E .3.2 E [X 2 ] E [Y2] < 2 y ' E [ y 2 ] E [ X 2 ] E [XY] -^E[X2]E[Y2] < E [XY]. aX + b) = E [X{aX + b)} .bY)2] = a2E [X2] .E [X] (oE [X] + b) = aE [X2] + bE [X] . [XY] + E [X2] E [Y2] By a similar set of steps the second inequality produces E [XY] < y/E[X2]E[Y2]. E [Y2] = 0.

Y)<l.8222) v For this data set there does not seem to be an especially strong linear relationship between height and weight.Y) < A/Var (X) Var (Y) (by Lemma 10.v / E [ X 2 ] E [ y 2 ] < E [XY] < y/E [X2] E [Y2] (E[XY]f <E[X2]E[Y2].1 illustrates this as well.Y) < Consider the covariance of X and Y.E [Y])])2 < E [(X .3 1.E [Y])2] = Var (X) Var (Y) Thus |Cov (X. Theorem 10.1) _x< Cov££L<1 " ^ V a r (X) Var (Y) D -l<p(X. . (Cov (X.Optimizing Portfolios 161 Therefore. . Proof.46944)(41. Y))2 = (E [{X .E [X])2] E [(Y . which is equivalent to the inequality. • Now we are in a position to prove the following theorem. Figure 10. Y) | < y^Var (X) Var (Y).2 Referring to the data in Table 10. A careful reading of the expression will avoid any possible confusion.3533.1 the correlation between height and weight is calculated as C 0 V ^F) ^ V a r (X) Var (Y) 2 7? -0. Example 10. since . which follows from the definition of correlation. /(1.E [X])(Y . Occasionally (E [XY])2 is written as E [XY]2 or even as E 2 [XY].V V a r ( X ) V a r ( Y ) < Cov(X. If X and Y are random variables then — 1 < p(X.

162 An Undergraduate Introduction to Financial Mathematics Wei ght 130 125 120 115 110 105 65 66 67 68 69 70 Height • • Fig.5 the concept of rate of return was defined to be the interest rate equivalent to this growth in value. then E[W]=E J2wi(l + Ri) ^2wi+^2wiE[Ri\. 10.1 A scatter plot of height versus weight for a sample of ten women who appeared as Playboy centerfolds. A portfolio may consist of n different investments. Suppose that an investor may invest an amount Wi for one time period and at the end of the time period. In Sec. If the rate of return for each of the n investments is treated as a random variable. 1. Thus the total accumulated wealth after one time period from all n investments is W = X^iLi Wi(l + Ri). For the ith investment the rate of return will be determined by Wi(l + Ri) = WiXi which implies Ri = Xi — 1. . the investment will return wealth in the amount of WiXi. We end this section with a more financial application of the concepts of covariance and correlation.

Ke»+CT*/2(j){w + a).i n + i e . Before concluding this section we will exercise our understanding of covariance and lognormal random variables to derive a technical result which will be needed in Sec.7) V2^a Jo x(x . Proof./ 1 1 r00 (10.K)+] = e2^+a2U(w with w = (/i — In K)/a. according to Eq. 10. Remember that the smaller the variance of a random variable. 10.Optimizing Portfolios 163 The variance in the accumulated wealth is.7.= / (e" " .4).^ 2 / 2 " 2 dx 1 ' x = -yL/ (X .6) + 2a) .M ) 2 / 2 ^ V27TCT JK 1 Z"00 + ^ = .K)+]J = . E \X(X .K)+) = E [X(X .0 n * . Later sections will expand on this idea and develop a method of selecting a portfolio of investments which minimizes the variance in the wealth generated. the smaller the spread about the expected value of that variable. Equation 10. The technical portion of the lemma is the derivation of Eq.6 follows from the definition of covariance. Lemma 10. where E [X(X .K)+] .# ) e .d " * . V27T J(\nK-n)/a . (10. (X .6.L .2 If X is a lognormal random variable with drift parameter ji and volatility a2 and K > 0 is a constant then Cov (X.E [X] E [(X . Minimizing the variance in the rate of return on a portfolio of investments makes the return more "predictable".K)+] . Var (W) = Var j ^ 1^(1 + ifc) J = Var ^WiRi n i=\ n 2—1 J n i=l j^i n = ^2 ^ 2 Var (i?j) + X X WiWjCov (Ri: i=l j^i Rj). (10.K^^e-*2'2 .

Upon reflection.+a /2 dz <*-')'/* dz The proof is complete if we let w = (fi — In K)/a.K)+] = roo poo 2 / e~(z-2-) /2 l(lnK~n)/a V27T J(lnK~n)/a 2 roo a/J. 9. In these cases the manager of a portfolio may have to find a surrogate for the security.+a ) •2(H+o-) E [X(X . If the hedge could be created with stock A in place of stock B then the Delta-neutral hedge is achieved when n = A as we saw in Sec. If insufficient shares of stock A can be purchased to create the hedge then the firm can investigate hedging their short position in the option for stock A with a long position in n shares of stock B. all of the discussion and examples assumed that a hedged position with respect to a particular option could be set up by taking a position in the stock or security underlying the option.164 An Undergraduate Introduction to Financial Mathematics The last equation is derived by making the substitution az = \nx — fi. what should n be when the hedge must be created with stock B? An optimal portfolio is one for which the variance in the value of the portfolio in minimized. If the changes in the two values are highly correlated then the portfolio manager will have more confidence in using the surrogate. • 10. a particular financial instrument will be a better or worse surrogate for a security depending on how the values of the security and the other instrument change. Therefore 2 e2(fj.2 Optimal Portfolios In the previous chapter on hedging. it may not always be possible to purchase (or sell) sufficient shares of the underlying security (or the option itself) to create the hedged position. However.2. The issue confronting the investment firm now is. Suppose an investment firm sells a European call option for stock A and wishes to create a Delta-neutral portfolio. The value of the portfolio consisting of a short position on a call option for .

The variance in the value of the portfolio is Var(V) = E [(CA .3 Utility Functions This section will focus on defining and understanding a class of functions which can be used as the basis of rational decision making. C 2 .nB. 10.n. The probability of outcome d coming about as the result of investing in the first product will be denoted pi for i = 1. .Optimizing Portfolios 165 stock A and a long position of n shares of stock B will be denoted V = CA . . The idea of selecting a portfolio by minimizing the variance of its value will be extended in later sections of this chapter.2nCov (CA. we must introduce the concept known as utility and describe its properties. B) + Var (CA) • Thus it is seen that the variance in the value of the portfolio is a quadratic function of the hedging parameter n. The reader should think of this set of outcomes as the union of the possible outcomes of investing in the first product and the possible outcomes resulting from investing in the second product. . Before returning to minimum variance analysis. C„}. B) = n 2 Var (B) . Suppose that an investor is faced with the choice of two different investment products. If outcome d cannot result from investing in the first product then. A small correlation between CA and B would indicate that stock B is a poor surrogate for stock A. Thus the minimum variance is achieved when Cov (CA.2nCov (CA. . 2 . . of course.B) =p(CA B) Var (OQ "" Var(B) ' V^rW Thus if the correlation between the values of the option on stock A and stock B were unity then the hedging parameter would be the ratio of the standard deviations in the value of the option on stock A and the value of stock B. . Suppose further that the set of outcomes resulting from investing in either of the products is {Ci. . .nB)2] -E[CAnB}2 = Var (CA) + « 2 Var (B) . .

The utility function provides a means by which two outcomes of unequal desirability may be compared.1). Outcome d is equivalent to receiving Ca with probability u(Ci) or receiving Cw with probability 1 — u(d).1].166 An Undergraduate Introduction to Financial Mathematics Pi = 0. This probability is called the utility of outcome C. This value of u will be the value of u(d). If u = 1 then the investor will surely participate in the random experiment because they will receive Ca with certainty. The values of u(d) for i ^ a. The utility function u(d) will have a range of [0. Thus the utility function is well-defined.Returning to the original task of deciding between two different investment products. i=l while the second investment is equivalent to Ca with probability n ^2qiu(d)- . to will lie in the open interval (0. To start. and they rank this as a more desirable outcome than d. As u decreases from 1 to 0 there will exist a value of 0 < u < 1 at which the investor will switch from the behavior of participating in the random experiment to taking outcome C.. They are assigned as follows. The investor imagines that they have the choice of receiving outcome d with certainty or participating in a random experiment in which they will receive the most desirable outcome Ca with probability u and the least desirable outcome Cw with probability 1 — u. the investor can then understand that investing in the first product is equivalent to receiving Ca with probability n ^2piu(d). u(Ca) = 1 and u{CJ) = 0.If « = 0 then the investor will definitely not participate in the random experiment since that would result in them receiving the least desirable outcome. The investor can rank the outcomes in order of desirability. with certainty. To each of the possible outcomes a utility can be assigned. with the most desirable outcome being Ca and the least desirable being Cw. It is assumed that the probability at which the behavior changes is unique. They are better off taking the sure outcome d. For the second product the values of the probabilities will be denoted qi. The rational investor once having decided to accept the sure outcome C. will not at a lower probability of receiving Ca decide to once again participate in the random experiment.

8) and otherwise will choose the second product.2 The extra utility received.2.1] we have interval (a. u(x + Ax) — u(x).Optimizing Portfolios 167 Therefore the investor will choose the first product whenever ^2piu(d) 4=1 > y ^ gjUJCj (10. b) if for every x. y G (a. This property is illustrated in Fig. More concretely these outcomes can be thought of as receiving differing amounts of money for an investment (some of which may be negative). 10. In other words u(x + Ax) — u(x) is a non-increasing function of x. In the remainder of this section categories and properties of the utility function will be explored.Ay) > A/(x) + (1 . b) and every /(Ax + (1 . Thus in general the utility function denoted u{x) is the investor's utility of receiving an amount x.X)f(y). A general property of utility functions is that the amount of extra utility that an investor experiences when x is increased to x+Ax is non-increasing. is a decreasing function of x. (10. In this section we have spoken of outcomes or consequences Cj of making an investment decision. 10. We will describe a function f(t) as u(t) u(x+Ax) u(x) X+Ax Fig.9) Graphically this may be interpreted as meaning that all the secant lines . concave on an open A € [0. Utility functions are specific (like personality traits) to each investor.

11) Making use of inequality (10. Theorem 10.4 / / fC2(a. (10. b).X)(y .fix) > (1 .10) and . If w = Xx + (1 . An investor whose utility function is concave is said to be risk-averse.9) produces the following pair of inequalities: / H . lie below the graph of /(£).9) will also be called concave. If / is concave on (a.3 For concave functions the secant line parameterized by A G [0.9). By definition of w. See Fig.f(w) < X(f(y) f{x)) f{x)) (10. The following result can be demonstrated regarding concave functions.x) > 0 y — w = X(y — x) > 0 (10.3.b) if and only if Proof.13) Through the use of inequalities (10. An investor with a linear utility function of the form u[x) = ax + b with a > 0 is said to be riskneutral. Let y E (a.12) (10. A utility function u(t).x = (1 . w . b) then by definition / satisfies inequality (10.1] will lie below the corresponding point on the graph of the function.b) f"(t) < 0 fora<t<b. obeying inequality (10.10) (10.168 An Undergraduate Introduction to Financial Mathematics U(t) u(y) u(x) Fig.13) and Eq. 10. 10. then f is concave on (a.X)y and if 0 < A < 1 then a < x < w < y <b.X)(f(y) f(y) .12) and (10.

s (w . .11) we obtain f(y) .x) < f(w) .Optimizing Portfolios 169 (10./(Ax + (1 . To prove the converse suppose / " ( x ) < 0 on (a. we can for any A £ (0.fix) w—x ~ From the inequality immediately above it is seen that / sf(y) ./ H ) y—w = /(y) .f(x) y-w (1 .A)(/(y) ./ M y—w < A(/(y) .A)y)) < Xf(Xx + (1 .*)m-f^-$~X)V) *^ + (! - A ^ A/(x) + (1 . + /(x) + (1 .b). ./ M y—w < / H ..(1 ..1) assign w = Xx + (1 — A)j/ and note that x < w < y.f(w) . Then for any x.X)y . b).x)f(v)-f& + fi-X)v) < f{Xx + ( 1 _ A ) y ) 2/ .y such that a < x < y < b./(x) ~ y—x _{l-X){f{y)-f{x)) w—x < w—x By the Mean Value Theorem [Stewart (1999)] there exist a and /? with x<j3<w<a<y such that /'(„) = Hv) ~ Hw) < J » ~ /( x ) y—w w—x Consequently a —p and in the limit we see that z—>y a —p = /' (/3 ).A)y f(x) + (A.A)/(y) < /(Ax + (1 . • . f(y) . .X)y) Xf(x) + (1 .Ax .A)y) which implies that / is concave on (a.A)(„ .

b) fori = 1.0. Since / is concave on (a.n. 2 .. the mean of the values of the function applied to a set of values of a random variable is no greater than the function applied to the mean of the values of the random variable. n with J27=i ^» ~ 1> then n / n \ ^AJtxO^/ i=l $ > ^ \i=l (10.2.1] for i = 1. . b). b) then / ( * 0 < f'{n){xi Therefore £ i=l . .A 3 .14) ) Proof. n.0. n and 5Z"=i -^ = 1 then a < [i < b.A4.30. Let ^ = Yl7=i ^iXi anc ^ n o t e that s m c e ^i € [0. Example 10. Theorem 10.23. .2. .. . [f'(n){xi n i=l M) + /(A*)]) n i=l = /'(Al)(At-M^Ai)+/(M) i=l = /(A0 V..07. then E [upQ] < u(E [X]). * Ai/^) < £ 2=1 (A.0.5 (Jensen's Inequality (Discrete Version)) Let f be a concave function on the interval (a.A 2 .3 Suppose f(x) = tanhx and let {A 1 . .t=l D Jensen's inequality may be interpreted as saying that for a concave function..170 An Undergraduate Introduction to Financial Mathematics In the next section we will make use of the following result known as Jensen's Inequality. 2 . The equation of the line > tangent to the graph of / at the point (/tx. and suppose Aj S [0. .. Thus a risk-averse investor prefers the certain return of C = E [X] to receiving a random return with this expected value.20. . /(/u)) is y = f'(n)(x — fi) + /(//). .A 5 } = {0. If u is a concave utility function.20}.A ) + /(A*) for i = 1.0. suppose xt G (a..1] for i — 1..

. The inequality in (10.5 then J2 W(Xi) = 0. Theorem 10. .8) indicated that if the first investment choice results in outcomes {C\. expected utility is the expected value of a utility function. • • •. f{a)).1] and let f be a concave function. oo). For the sake of compactness of notation let a — J0 <f>{t) dt. If Xi = i2 for i = 1. and let y = f'(a)(x — a) + f(a).. .928431 < 1 = / j J2 >*xi) There is also a continuous version of Jensen's Inequality. then fW))dt<f(J <P{t)dt\- (10. + f(a) dt f Jo f(a)dt / ' ( a ) j <j)(t)dt~ [ f'(a)adt+ Jo Jo / ' ( a ) a . n and the second investment choice .4 Expected Utility As the name suggests. Returning to the previous discussion of choosing between two possible investment instruments. we see that the investment with the greater expected utility is preferable.Optimizing Portfolios 171 The reader can verify that f(x) is concave on (0. .2.6 (Jensen's Inequality (Continuous Version)) Let 4>{t) be an integrable function on [0.. C„} with respective probabilities pi for i = 1 2 . the equation of the tangent line passing through the point with coordinates (a. Cg. Since / is concave then fWt))<f'(a)m)-a) which implies that f im))dt< o j f'(a)(<f>(t)-a) Jo + f(a)./ ' ( a ) a + /(a) f(f mdt a 10. .15) Proof.

• • •. fti}. This illustrates the concept known as the certainty equivalent which is defined as the value C of a random variable X at which u(C) = E[u{X)\. The investor is risk-averse with a utility function u(x) = x — x2/25. The expected utility for investment A is 1 1 1 / 102 The expected utility for B is u(M) = M . if the coin lands heads up the investor receives $10.172 An Undergraduate Introduction to Financial Mathematics produces the same outcomes but with probabilities {ft. • • • . • • • . Example 10.49. qn} then a rational investor would select the first investment whenever n i=l n i=l Suppose the random variable X is thought of as the set of outcomes with probabilities {Pi.5 An investor wishes to find the certainty equivalent C for the following investment choice: .P2. The rational investor will select the investment with the greater expected utility.49 in the previous example. 2 Thus investment A is preferable to B whenever M < $3.ft>.4 vestments" : An investor must choose between the following two "in- A: Flip a fair coin.M 2 / 2 5 .25M + 75 > 0 25-5^13 w — > M. otherwise they receive nothing. then the investor will choose the second investment.Pn} while the random variable Y represents the outcomes with probabilities {ft. B : Receive an amount $M with certainty. Thus the investor will choose the coin flip whenever M2 3>M 2 --2T M . If M > $3. Example 10.Then the first investment is preferable whenever E[u(X)]>E[u(Y)}. 52.

Thus for an investment of ax the random variable representing the investor's financial position after the conclusion of the investment is Y ( x(l + a) with probability p. 10. The allocation proportion a will be optimal when the expected value of the . The design of investment A specifies that 0 < Y < X < 10. The investment is structured such that an allocation of ax will earn ax with probability p and lose ax with probability 1 —p. It is clear from the expression under the radical that the certainty equivalent is real only when X2 + Y2 < 625/48.1] be the proportional they invest. Assuming that the investor may use any proportion of this capital. all-or-nothing decision. In reality an investor may choose to split investment capital between two (or more) investments. if the coin lands heads up the investor receives 0 < X < 10. Figure 10.Optimizing Portfolios 173 A: Flip a fair coin. Thus the certainty equivalent can be thought of as a surface plotted in cylindrical polar coordinates over the region where 0 < r < -?4= and 0 < 9 < 7r/4. In this section we will explore the portfolio selection problem.4 illustrates the certainty equivalent as a function of X and Y. B : Receive an amount C with certainty. the certainty equivalent and payoffs of investment A must satisfy the equation l (x X \Y Y2 \-C 25-y/625-48(X2+Y2) - C - C 2 -25C+12(X2 + Y2) = 0 C. with probability 1 — p. We begin with a simple example.5 Portfolio Selection In the previous sections we have treated the investor's choice between two different investments as a binary. \x(l — a). otherwise they receive 0 < Y < X. the task of allocating funds in an optimal fashion among several investment options. Suppose an investor has a total of x amount of capital to invest. Assuming the investor's utility function is the same as in the previous example. let a £ [0.

then the expected value of the investor's utility is maximized when __" 1+ a tL — o. The domain of interest in the XV-plane is drawn as a shadow below the surface.x(l .(1 .a)) do. 1 — a: which implies a — 2p~ 1.p)u'{x(l .6 If u(x) = lax. 0 = pu'(x(l + a)) .p)u(x(l . utility is maximized.p)u'(x(l ~ a)) The critical value of a which solves the last equation above will correspond to a maximum for the expected value of the utility as long as the utility function is concave.a)) -£-E [u(X)\ = pxu'(x(l + a)) . E [u{X)] . 10. E x a m p l e 10.174 An Undergraduate Introduction to Financial Mathematics Fig.pu(x(l + a)) + (1 .4 The certainty equivalent in this example is a section of an ellipsoid. .

5 0. 2 .60.2. 10.8 0. when no investment is made. The return from investment i will be denoted Wi for i = 1. With this simple example understood. Under these assumptions the Central . .4 0.5 The expected value of utility for a risk-averse investor allocating 100(2p— 1)% of their capital. Suppose that an amount of capital x may be allocated among n different investments. xn) will be used to represent the portfolio of investments. E[u(X)] 0.2 0. . n such that (1) 0 < Xi < 1 for i = 1. 2 . . . . Throughout this section it will be assumed that n is large and that the Wi's are not too highly correlated. then E [u(X)] is maximized when a = 0.4 0. 10. The portfolio selection problem is then defined to be that of determining Xi for i = 1.Optimizing Portfolios 175 Thus if p > 1/2 the investor should allocate 100(2p — 1)% of their capital. 2 .3 0. A proportion xi will be allocated to security i for i = 1.5.n. The vector notation (x\. n. and (2) £ I U ^ = 1.7{ 0. . . . we are now ready to consider a more general situation. . . If 0 < p < 1/2.e. . n.2 0. . . • • •. i.1 0. X2. and which maximizes the investor's total expected utility E [u(VF)].6 Fig. A curve depicting the expected value of the investor's utility as a function of a and p is presented in Fig. where n W = xJ^XiWi.

Security E [rate of return] Var (rate of return) A 0. • • •. Vn). then Y is a lognormal random variable (see Sec. According to Lemma 3.E [e-bw] .24 . Assuming that W i s a normally distributed random variable then — bW is also normally distributed with E [-bW] = .6 E [W] Consequently.16 0.1.20 B 0. xn) will be preferable if E [X] > E [Y] and Var (X) < Var (Y).18 0.005 and that an investor wishes to choose among an infinite number of different portfolios which can be created by investing in two securities denoted A and B.6). • • •. The allocation of investment funds can be driven by the objective of maximizing the expected value of the investor's utility function. Making the assignment Y = e~bw. X2. E [u{W)\ = E [1 . give rise respectively to returns X and Y then the portfolio represented by (xi. xn) and (yi. The investor has $100 to invest and will allocate y dollars to security A and 100 — y dollars to B. (x\. The following table summarizes what is known about the rates of return on the hypothetical securities A and B.y2. 3.X2. suppose b = 0.5 implies that W is a normally distributed random variable. To extend this discussion and make it more concrete numerically.176 An Undergraduate Introduction to Financial Mathematics Limit Theorem of Sec.e~bx where b > 0. E [Y] = E [e-bW] = which implies that e-mw]+b 2 v^(w)/2 Thus the expected utility is maximized when E [W] — &Var (W) /2 is maximized. Suppose an investor's utility function is given by u(x) = 1 . and Var (-bW) = 62Var (W). • • •. 3.e'bw] = 1 . The reader is asked in exercise 12 to show that this utility function is concave and monotonically increasing. If two different portfolios.

20)(0.0172y + 116. Once again W will represent the wealth returned.a).2195. Without loss of generality it is assumed that the investor has an infinitely divisible unit of capital to invest and will invest a fraction a G [0.yf .476 Var(W) « 276.16y + 0.6 Minimum Variance Analysis Building on the ideas contained in the previous section we consider the case of an investor allocating a portion of their capital between two potential investments.y)(0.04y2 + 0. E [W] = 100 + 0.04y2 + 0. (10.Optimizing Portfolios 177 Assume that the correlation between the rates of return is p = —0.35.yf .0.The variances of these rates of return will be denoted cr\ and o\.35) = 0. The covariance in the rates of return is assumed to be c.442296 In the next section we will generalize and extend these ideas to situations in which investment capital is partitioned n ways.16) One measure of an optimal portfolio of investments would be the one with the least variance in the returned wealth.1] in one security and 1 — a in a second security.0576(100 .0.0.02y .000328j/ 2 + 0.24) 2 + 2y(100 .y) Since the optimal portfolio is the one which maximizes E [W] — 6Var (W) jl then we must maximize 118 .0336y(100 .20) 2 + (100 . The rates of return of the two securities are respectively R\ and i?2.56.02?/ Var (W) = y 2 (0.0336y(100 .y) = 118-0. Differentiating Var (W) with .18(100 .y) 2 (0. For this choice of y E[W] « 117. 10.y)) = -0.0025(0.24)(-0.0576(100 .0. This occurs when y sa 26. Thus the variance in the wealth returned from the investments is Var (W) = o?a\ + (1 .049 E [u{W)\ « 0.afal + 2ca(l .

.2c The denominator of a* is the covariance of . a = 1. Under this assumption c = 0 and the critical value of a at which the variance in the returned wealth is minimized reduces to a* = " n2 °l 2 -i- = l °l . a „ ) . This occurs whenever c < min{(Tj. If c < 0 then this condition is always satisfied.?. minimum variance portfolio described by the allocation vector {ai.a)a\ + c(l .cr2}.1]. This is not due to coincidence. it is assumed that a\ +<rf > 2c. • • • . the minimum variance will occur when a = 0. The optimal. According to the Extreme Value Theorem [Stewart (1999)].178 An Undergraduate Introduction to Financial Mathematics respect to a and solving for the critical value of a produces 0 = 2 (aof .2a)) * a = o\ + o\.. Var(W)| Q = 0 = a22 Var(W)|Q=1=<r? °2 ~c Consider the special case when the rates of return of the two investments are uncorrelated. The parameter a* must also fall within the closed interval [0.(1 . is the . This condition is also sufficient to guarantee that the variance of the wealth in Eq.ct2.Ri — R2 and thus is non-negative. In order for the critical value of a to be defined. it is in fact a pattern seen in the more general case of allocating a unit of investment among n different potential investments. or a — a*. T h e o r e m 10. n subject to the constraint that ct\ + a.2.16) will have a global minimum. Suppose the rate of return of security i is a random variable Ri and that all the rates of return are mutually uncorrelated. _2 i l ' °1 + a2 -^ + -^ °1 °2 There is an appealing simplicity and symmetry to this critical value. (10.7 Suppose that 0 < on < 1 will be invested in security i for i = 1. + • • • + an = 1...

• • . . . ai € [0. These equations are equivalent to respectively: 2aiaf = A for i = 1. . .n*7' i for i = l . where of = Var (Ri). Since the rates are return are uncorrelated then the variance in the returned wealth W is Var(W0 = 5 > ? a ? ... . . . an. 2 . Proof.. . 2 .2. n..Minimizing the variance subject to the constraint is accomplished by use of Lagrange Multipliers [Stewart (1999)]. and n i=l Solving for ai in the first equation and substituting into the second equation determines that A E 9 n j _ 3 = 1 a? Substituting this expression for A into the first equation yields l ^3 = 1 a? . • —5" The reader can confirm that for each i. n . a. n. i=l and is subject to the constraint that 1 = Y^i=i ai. A) of the set of simultaneous equations: n The symbol V denotes the gradient operator..Optimizing Portfolios 179 one for which — for i = 1. .2. This technique states that Var (W) will be optimized at one of the solutions (c*i.1].

r(w) = E[#(w)] cr (w) = Var(i?(w)). Wi.180 An Undergraduate Introduction to Financial Mathematics The previous discussion can be generalized yet again to the situation in which the portfolio of securities is financed with borrowed capital. Theorem 10. As before.3 Assuming the rates of return on the securities are uncorrected.20). (10. u>2.3 will be used in the following result known as the Portfolio Separation Theorem. . rn—r w. 2 . . the optimal portfolio with an expected return of unity which minimizes the variance of the portfolio is I r\—r r^ — r w.7 and is left as an exercise. The existence of the optimal portfolio w* provided by Lemma 10. The proof of this lemma is similar to the proof of Theorem 10.20) ^? <T? where ri = E [Ri] and of = Var (Ri) for i = 1..18) (10.8 (Portfolio Separation Theorem) Ifb is any positive scalar. For simplicity's sake we will assume this is a one-period model in which investments are purchased by borrowing money which must be paid back with simple interest of rate r. . . .. The return on the portfolio after one time period is ^( ) = Yl *( + i)-( + )J2 2=1 w n w 1 R l r n Wi = J2 w*(R* -r)2=1 n (10-17) 1=1 The expected rate of return on the portfolio and the variance in the rate of return on the portfolio are functions of the vector w and are defined to be respectively.19) In this situation.. the variance of all portfolios with expected rate of return equal to b is minimized by portfolio few* where w* is described in Eq. \ w) w*= j= <7 = i1 "-j T2 "-j 2^j=i 2->j = m "-f 2->j=i (10. or to any other prescribed value.. Lemma 10. Ri will represent the return on investing Wi in the ith security. • • •. 2 (10. . wn do not have to sum to unity. Let w = (wi. . wn) represent the portfolio of investments. the borrowed amounts W\. . n. .

If the rate of return of the security is Rs and x £ [0. Suppose x is a portfolio for which r(x) = b. o-2 (6w*) = & V ( w < f >< ( i x ) r2 V <r2(x) (by Lemma 10.x)E [rf] + xE {Rs} = (1 — x)rj + xrs r = rf + (rs -rf)x (10. which implies the expected rate of return for this portfolio is E [R] = (1 . Thus if the expected rate of return of the security exceeds the risk-free rate of return.1] is invested in the security while 1 — x is placed in the risk-free investment. then -r(x) = r I . its expected value is this same constant value.3) D As a consequence of the Portfolio Separation Theorem all expected rates of return on portfolios can be normalized to unity when determining the optimal portfolio. For the portfolio w*. By assumption the rate of return on the risk-free investment is guaranteed. the slope of the line is positive and the expected rate of return of . The risk-free interest rate will be symbolized by rf. The line has its r-intercept at 77 and slope rs — rf. The expected rate of return on the portfolio is a linear function of x. As a special case of the preceding discussion the investor may wish to divide their portfolio between a risk-free investment such as a savings account and a "risky" investment such as a security. has a variance of zero.e. i. the rate of return on the portfolio is R = (1 — x)rf + xRs.21) (by exercise 15) To achieve a more compact notation we have set r = E [R] and rs = E [Rs]Since the risk-free return is a constant.Optimizing Portfolios 181 Proof. (by exercise 15) Thus we see that £x is a portfolio with unit expected rate of return.x I = 1.

This is known as the equation of the capital market line. Rs) a2 = x2a2s (10.182 An Undergraduate Introduction to Financial Mathematics the portfolio increases with x. For a fixed level of expected return. r = rt -\ -a Once again. r r = rf + J£^lLa (10.23) The capital market line is illustrated in Fig. Referring to the illustration of the capital market line in Fig. risk-adverse investors will want to minimize the variance (or standard deviation) in the rate of return. Before moving on. for a rate of return above 77 it is not feasible to have zero variance in the rate of return.21) and to derive a formula relating the expected rate of return and the standard deviation of the rate of return. A fixed level of return corresponds to a horizontal line in the or-plane. If the idealized single security of the previous discussion is replaced with an investment spread evenly throughout the securities market then the same form of linear relationship is determined. The phrase "as is feasible" is used because it may not be possible to achieve a given rate of return with zero standard deviation. otherwise it decreases.cc)2Var (77) + x 2 Var (Rs) + 2z(l . its variance is zero and the covariance with the rate of return on the security also vanishes.22) can be used to eliminate the parameter x from Eq. Thus a risk-adverse investor would prefer a portfolio for which the standard deviation in the rate of return is as far to the left on the horizontal line as is feasible. 10. for a . The reader will note that since the rate of return on the risk-free investment is constant. 10. Equation (10.6. The standard deviation of the return on the portfolio is also an increasing linear function of x.6. but the symbols TM and UM will be used in place of rs and as respectively.22) where Var (R) = a2 and Var (-Rs) = a%. if the expected rate of return on the security exceeds the rate of return of the risk-free investment. Thus an investor willing to accept a wider variation in the return on funds invested can expect a higher rate of return. the expected rate of return of the portfolio increases with the standard deviation in the return on the portfolio. Similarly. The variance in the rate of return on the portfolio is Var (R) = (1 .x)Cov (77. the reader should reflect on the or-plane. (10.

for a fixed a not all levels of r are achievable. an investor should prefer the highest expected return. 10. the Capital Asset Pricing Model (CAPM). fixed standard deviation in the rate of return. which implies the expected rate of return on the portfolio and the variance . This mathematical model will relate the return on the investment in an individual security to the return on the entire market.6 The capital market line illustrates the linear relationship between the standard deviation in the return on a portfolio and the return on the portfolio. An investor will invest a portion x of his wealth in investment i and the remainder 1—x in the market. The rate of return on the portfolio is given by R = xRi + (lx)R_M. Thus along a vertical line in the or-plane an investor should prefer a portfolio as high on the line as is feasible. They are similar to those done when determining the optimal allocation of a portfolio between two securities. Finally we arrive at a discussion of the main topic of this chapter. In this way an investor can weigh the return on the investment against the risk involved in the investment. The expression Ri will be the return on investment i and RM will denote the return on the entire market. Again.Optimizing Portfolios 183 Fig. These ideas could be extended a great deal and the interested reader is encouraged to consult [Luenberger (1998)]. This time the second security is the entire market. The following calculations will seem familiar to the reader.

= 2 2 2 x a (10. the parametric curve intersects the capital market line. When x = 0.M&il&M is commonly denoted (3i and is referred to as the b e t a of security i.23) to obtain the slope of the capital market line we have TM -rf _ dr da dr I dx la:—0 da I dx \x=0 (JM n . and rj. To simplify the notation we will make the following assignments: r = E[R].x) a 2 2 + 2x(l . 1 to denote the .)Cov (Ru RM).x)rM o.rf yields n-rf a = Pi. °M Solving the last equation for r» -.M i t 77).26) CM The expression Pi.24) M + (1 . E [R] = xE [Ri] + (1 . then for some value of x near zero the parametric curve would lie above the capital market line and hence represent an infeasible expected return. Equation (10.RM) correlation of Ri and RM • Here there is little chance of confusion.)2Var (RM) + 2x(\ .x)piM<7i<?M- (10.25) can be thought of as the parametric form of a curve in the or-plane.rM W7-.184 An Undergraduate Introduction to Financial Mathematics in the expected rate of return are respectively.25) Equations (10. n = E[Ri]. If the intersection were transverse rather than tangential. Using Eq. In fact the parametric curve must be tangent to the capital market line at x = 0.24) and (10. Thus we can obtain a relationship between r^. (10.Furthermore we will replace Cov-(R^RM) by its slightly more compact equivalent Pi^u^i^M-1 Thus the equations above become r = xri + (1 .a. corresponding to the situation in which the entire portfolio is invested in the market. Without further information we cannot assume that the rate of return for investment i is uncorrelated from the rate of return for the market. (10.a. o2 = Var(i?). rjvf. rM = E[RM]. and o2M = Var(i?M).26) can be interpreted as stating that the excess rate of return of security i above the risk-free interest rate is This is a small abuse of notation since we normally use P{RI.x)E [RM] Var (R) = ir2Var (Ri) + (1 . of = Var(Ri).

When the correlation is negative. 22. For a given level of variance in the rate of return. The maximum and minimum value of expected rate of return will occur when ^ = ^ (10-27) ax ax subject to the constraint expressed in Eq.78%.M ~ !K 2 ( J M + K 2 + ^M ' M (10. the expected rate of return on the market is 9% per year.05 + 4.2piiM0'i(TM)0'2 (10. or expressed as a percentage return.Optimizing Portfolios 185 proportional to the excess rate of return of the market above the risk-free interest rate.MaiaM 2pi. and the standard deviation in the return on the market is 15% per year. excess positive expected return for the market will imply excess negative return for security i. Thus once again we turn to Lagrange Multipliers to solve this constrained optimization problem.09 . (0.28) Pi. The reader will be asked to verify in the exercises that the solution to this set of equations is A= ± °M ~ °f + [ CT n ~™ = .44444(0. the investor will expect the highest expected value of the rate of return.M0'iCrM)0' 2 M ~~ ^iP2i. The variance in the rate of return of a portfolio can be thought of as a measure of the risk in the portfolio.22777.0.7 Suppose the risk-free interest rate is 5% per year.25). the excess expected return will be zero. The proportionality constant is /%. For the case in which the rate of return on security i is uncorrelated with the rate of return on the market. Example 10.05) = 0. excess positive expected return for the market implies excess positive expected return for security i. If the covariance in the expected returns on a particular stock and a the market is 10% then P 0 = —: = 4 44444.M(TiCTM . (10.M°iaM 2-Pi. When the correlation is positive.15)2 Therefore the expected rate of return on the stock will be r = 0.29) ^Pi.

M)\M[M ~ l)aiaM a + (°f +aM~ ^Pi. Suppose the rates of return have a correlation of p^M = 0. In this section the portfolio will consist of positions in a security and a European call option on the security. The reader should not be confused by the portfolio. The investor will purchase a units of the security and b units of the call option.07 with standard deviation CTJ = 0. Example 10.30) IPiMOiOM)** 2 i r + °M ~ 1 2 'Pi.—^ a nnon\ (10.05 with standard deviation OM = 0. i.22 then the maximum rate of return can be found via the method outlined above. The curve depicting r and x as parametric functions of a is shown in Fig. .25 and the expected rate of return for the market is ru = 0.186 An Undergraduate Introduction to Financial Mathematics Under these conditions the maximum expected rate of return is r = rM zp (n . Thus we will assume that ai > GMAs two special cases. According to Eq. If the rate of return on investment i is anti-correlated with the rate of return on the market. then r = rM (rt -rM)((?M • ±0") .7.0650248. 10.e.rM) (pi.30).57. when piiTn = 1.751242 or when slightly more than 75% of investment capital is placed in investment i. we will explore the return on the portfolio when the rates of return on the market and investment i are perfectly correlated.< 4 ) —. If necessary.M(Ji(JM 0* .8 Suppose the expected rate of return for investment i is Ti = 0. If an investor is willing to hold a portfolio with a standard deviation in its rate of return at a = 0.20. the reader may wish to review Chapter 5 before proceeding. (10. This occurs when x = 0. 10.7 Mean Variance Analysis This chapter concludes with a discussion of the stochastic process approach to determining the expected rate of return of a portfolio.MaiaM i +(Tll _ In most scenarios the variance in the rate of return for investment i will exceed the variance for the market.M<TiCrM . the maximum rate of return on the portfolio is 0. In this case r = rM (n -rM){<JM ±o-) Oi ~ CM .

(aS(0) + bC). At the expiry date T the positions in the portfolio are zeroed out. and the expiry date is T. and the proportion of capital invested x. At time t = 0 the investor devotes resources in the amount of aS(0) + bC to create the portfolio.3) then dS(t) /i + — ) S(t) dt + aS(t) dW(t). We will assume the present value of the security is S(0).2 0. By Ito's Lemma (Lemma 5. The present value of any gain or loss is then R = e~rT (aS{T) + b(S{T) . stochastic process with drift parameter fi and volatility a.6 0.07 0. For example.4 0. the value of the option is C. 10. a short position in the call option can be offset with a long position in the security. Likewise.09 0.8 Fig. The riskfree interest rate is r.Optimizing Portfolios 187 0. the investor might take a short position in the security and hedge this position with a long position in the call option.08 0. The reader will recall that this means that the logarithm of the value of the security follows a Wiener process of the form: d(lnS(t)) =fidt + (rdW(t). . where either a or b could be positive or negative. Suppose the value of the security obeys a Brownian motion-type.7 This curve shows the relationship between the rate of return r. the strike price is K.K)+) .

+ nT)/a\/T.bC. E [(S(T) .K)+] . .24) the expected value and variance of In S(T) are respectively E [In S(T)} = fiT and Var (In S(T)) = a2T.1.aS(0) . In order to determine a closed form expression for E [(S(T) — K)+] we will make use of Corollary 3. Thus by Lemma 3. S{T) is a lognormal random variable with parameters In5(0) + fiT (the drift parameter) and CTVT (the volatility parameter)..(\n. Provided that r < n+a2/2 the expression e-rTE[S(T)} = S{0)e(-r+»+a2/2)T > 5(0).bC (aE [S(T)] + bE [(S(T) .K)+) .K<j> (w).188 An Undergraduate Introduction to Financial Mathematics Using the linearity property of the expected value.aS(0) . Thus the expected value of the rate (10.2.aS(0) . Thus according to Corollary 3.bC] [aS(T) + b(S(T) .K)+]) .31) (S(0)e^ ^ +S )T (a + bcf>{w + aVf)) . (5.bK<P(w)^j .a5(0) . the expected return is E[R]=E [e~rT (aS(T) + b(S{T) .K)+1 = 5(0)^+^)^ (ln5(0)+g-lng T. rT = e-rTE = e~ According to Eqs. E[S(T)] = 5(0)e^ + f f 2 / 2 ) T .S{0)+nT-\nK + ^ ~ V WT (HS(0)/K) + fiT CT\/T) = S(0)e^+a2WT<f> (w + where w = (\n(S(0)/K) of return is r = E[R] = erT .23) and (5.2.bC The level sets of the expected return are lines in the afo-plane.

aS(0) .Optimizing Portfolios 189 The variance in the rate of return can be calculated as Var (R) = Var (e~rT (aS{T) + b(S{T) .bC) = Var (e~rT (aS{T) + b(S{T) K)+)) K)+) = e. (S(T) . but complete. the complicated.1 we have Var (S(T)) = S ^ O ^ 2 " * ^ (e^T According to Corollary 3.4. Referring to the second part of Lemma 3. .K)+) = S2(0)e2^+"2)T <l>(w + 2aVT) +a 2)T .2KS(0)e^ ^ (j)(w + aVf) w+a + K2cj)(w) Lemma 10. (S(T) .5% . at long last.9 Suppose a share of a security has a current price of $100 while the drift parameter and volatility of the price of the security are respectively 0. Var ((S{T) .K)+)) .K)+) .32) b2(s2(0)e2^+^T(t>(w+2aVf)-2KS(0)e^+°2WT(p(w+aVf) K2(t>{w)-(^S(0)e(-fl+a2/2)T(j)(w+(jVT)-K(t){w)y^ WT (4>(w)-(t>{w+(jy/f)) + 2ab(KS(Q)e^+a2 - S2{0)e2^+a2)T((l){w+aVf)-4>(w+2(TVf)^S) The level sets of Var (R) are parabolas in the afo-plane.08 and 0.2 implies that Cov (S(T).2 r T Var (aS(T) + b(S(T) . expression for Var (R). V&Y(R)=e-2rT(a2S2(0)e^+^T(e°2T-l) + + (10.21 per annum.l) . The risk free interest rate is 3.K)+) =KS(0)e^+a2/2)TU{w)-4>(w+aVf)') S2(0)e2^+ff2)TU(w+aVT)-^{w+2aVf)^ Putting these last three results together yields. E x a m p l e 10.K)+) = e~2rT (a 2 Var (5(T)) + 62Var ({S(T) + 2abCov (S{T).

The value of a European call option for a strike price of $102 with an exercise time of 12 months on the security is $9.6.58208a:. We look for a point in the second or fourth quadrant since typically x and y must be of opposite sign due to the fact that the investor will take opposite positions in the security and the option.8 The expected return of the portfolio is positive above the diagonal line shown in the density plot. An investor can use the preceding analysis to determine that the expected rate of return on a portfolio consisting of a position x in the security and a position y in the call option is given by Eq. (10. If the investor Fig.383%. A density plot in Fig. 10.8 shows the return in various subsets of the plane. The line of zero return is given by the equation y — -1.31) as E [R] .93489a) + 4. Thus at a point in the second or fourth quadrant above this line the expected return will be positive.075. 10.190 An Undergraduate Introduction to Financial Mathematics annually. .

353511 and the minimum variance is Var (R) = 37. the CEO's golf handicap.2 + 748. 05/31/1998.1. . and the corporation's stock rating.891 .8 Exercises (1) Using the definition of covariance and variance prove the first two statements of Theorem 10.902y2 Rather than use the method of Lagrange multipliers to find the minimum of Var (R) we will take the more elementary approach of solving the equation 2 = 6. Y).Optimizing Portfolios 191 decides that a return of E [R] = 2 is desirable they will then select x and y so that the variance in the return is minimized over all portfolios for which E [R] = 2. According to Eq.93489a.7. 10. E [Y2] = 0. It lists the name of a corporate CEO. and a rating of the stock of the CEO's corporation. Under these conditions y = —0.3834y for y and substituting into the expression for variance.9349.101. pg. (10. E [X2] = oo.1 for the cases in which E [X2] = 0.511844.2 which is minimized when x = 0.32) the variance in the return is Var (R) = 515.802x1/ + 306. 1. Section 3. If X and Y are independent random variables. + 99.421a. + 4.571a. show this formula reduces to the result mentioned in Theorem 2. In this case we get the quadratic expression Var (R) = 63. (3) Fill in the remaining details of the proof of Lemma 10. or E [Y2] = oo (4) The data shown in the table below was originally published in the New York Times. (2) Show that for any two random variables X and Y Var (X + Y) = Var {X) + Var (Y) + 2Cov {X. This is yet another constrained optimization exercise. the CEO's corporation.0737a. Determine the covariance and correlation between the CEO's golf handicap.

9 7. Shoemate James E. James E. The company hedges its position by buying a different security for which the standard deviation in value is ar) = 0.4) is sometimes called the arithmetic mean.86. Perrella William P. The correlation between the values of the two instruments is p (O.1 10.33) X The geometric mean is defined as P(X) (10. What is the optimal ratio of the number of shares of security purchased to the number of options sold so that the variance in the value of the portfolio is minimized? (6) Which of the following utility functions are concave on their domain? (a) (b) (c) (d) u(x) u(x) u(x) u(x) = = = = lnx ln(a.045.6 17. The standard deviation on the price of the call option is ao = 0.2) (lnx) tan^ 1 x (7) The mean of a discrete random variable as defined in Eq. Cayne John R.192 An Undergraduate Introduction to Financial Mathematics Name Terrence Murray William T.6 12.1 11 12.9 12.6 16. McCoy Frank C. Stiritz Corp.1 10. (2.34) Use Jensen's inequality (10. Larsen Paul Hazen Lawrence A. Bossidy Charles R. McColl Jr. D) = 0. Fleet Financial Sprint Nationsbank Bear Stearns Amer.8 13 Rating 67 66 64 64 58 58 55 54 54 51 49 49 48 (5) Suppose that a company sells a European call option on a security.6 10.14) to show that H < Q < E [X] and that .037. The harmonic mean is defined as H = E x l M 1 (10. Stafford John B. Esrey Hugh L. Herringer Ralph S. Home Products Banc One Transamerica Johnsonfe Johnson Wells Fargo Allied Signal Bestfoods Ingersoll-Rand Ralston Purina Handicap 10.6 10.

41 0. Suppose you must choose between receiving an amount C with certainty or playing a game in a fair die is rolled. You may assume that all the random variables are positive.Optimizing Portfolios 193 (8) (9) (10) (11) (12) equality holds only when X\ = X2 = • • • = Xn. (a) Show that u(x) is concave. The expected rate of return for the second investment is r-i = 0.15) for f(x) = tanha. Verify inequality (10. Assume that your utility function is f(x) = x — x2/2. determine the proportion of the portfolio which will be allocated to each security so that the variance in the returned wealth is minimized. Security A B C D E a2 0.08 with a standard deviation in the * rate of return of o\ =0. What is the optimal amount of money to place into each investment? (14) Suppose that an investor will split a unit of wealth between the following securities possessing variances in their rates of return as listed in the table below. Assuming that the rates of return are uncorrelated. If a prime number results you win $15. Assume that your utility function is f(x) = x — x 2 /50. or (b) receiving an amount C with certainty. Suppose a risk-averse investor with utility function u(x) = In a. and <j>(t) = t.33 . Find the certainty equivalent for the choice between (a) nipping a fair coin and either winning $10 or losing $2.09. The amount of capital not invested will earn interest for one time period at the simple rate r.16 0.13 with a standard deviation in the rate of return ofCT2= 0.26. (b) Show that 0 < x\ < X2 implies u(xi) < u(x2)- (13) Suppose an investor whose utility function is u{x) = 1 — e~x/100 has a total of $1000 to invest in two securities. but if a non-prime number results you lose that amount of money.03. will invest a proportion a of their total capital x in an investment which will pay them either 2ax with probability p or nothing with probability 1—p. The expected rate of return for the first investment is 7 1 = 0.24 0. The correlation in the rates of return is p = —0.27 0. What proportion of their capital should the investor allocate if they wish to achieve the maximum expected utility? Consider the function u(x) = 1 — e~bx where b > 0.

11 0. If the covariance in the expected returns on a particular stock and the market is 15%.25 per annum.7 to determine the positions of the optimal portfolio with an expected return of $10.17 a2 0.18) and (10.15 0.65% per year. (19) Suppose the risk-free interest rate is 4. the expected rate of return on the stock market is 7.27) are given by the values of A and x in (10.13 0.29). Security A B C D E r 0.194 An Undergraduate Introduction to Financial Mathematics (15) For any real scalar c show that for r(w) and er2(w) as defined in Eqs. Use the mean-variance analysis of Sec.33 (18) Verify that the solutions to the set of Eqs. determine the expected rate of return on the stock.19) the following conditions hold. The risk free interest rate is 5.3.25) and (10. Find the value of 3-month European call option on the security with a strike price of $86.13 and 0.16 0. (10. and the standard deviation in the return on the market is 22% per year.41 0. .75% per year.12 0.28) and (10.5% annually. (10. r(cw) = cr(w) <T 2 (cw) = C 2 <T 2 (w) (16) Prove Lemma 10. (20) Suppose a share of a security has a current price of $83 while the drift parameter and volatility of the price of the security are respectively 0.24 0. 10. (17) In this exercise we will extend the work that was done in exercise 14 by determining the optimal portfolio which minimizes the variance in the return under the condition that the portfolio is financed with money borrowed at the interest rate of 11%.27 0.

82 49.1 45.14 44.Appendix A Sample Stock Market Data In this appendix are end-of-day closing stock market prices for Sony Corporation stock.44 44.49 47.97 46.0 43. 2001 until August 12.85 42.21 44.95 47.siliconinvestor. 2002. The data were collected from the website www.7 45.01 42.55 45.86 44.71 44.3 48.65 46.33 46.59 195 .5 42.com and cover the one year period of August 13. Date Aug-12-2002 Aug-9-2002 Aug-8-2002 Aug-7-2002 Aug-6-2002 Aug-5-2002 Aug-2-2002 Aug-1-2002 Jul-31-2002 Jul-30-2002 Jul-29-2002 Jul-26-2002 Jul-25-2002 Jul-24-2002 Jul-23-2002 Jul-22-2002 Jul-19-2002 Jul-18-2002 Jul-17-2002 Jul-16-2002 Close 42.

1 50.69 57.45 53.27 54.11 58.98 53.36 55.06 51.78 52.01 54.41 50.2 50.3 48.196 An Undergraduate Introduction to Financial Mathematics Date Jul-15-2002 Jul-12-2002 Jul-11-2002 Jul-10-2002 Jul-9-2002 Jul-8-2002 Jul-5-2002 Jul-3-2002 Jul-2-2002 Jul-1-2002 Jun-28-2002 Jun-27-2002 Jun-26-2002 Jun-25-2002 Jun-24-2002 Jun-21-2002 Jun-20-2002 Jun-19-2002 Jun-18-2002 Jun-17-2002 Jun-14-2002 Jun-13-2002 Jun-12-2002 Jun-11-2002 Jun-10-2002 Jun-7-2002 Jun-6-2002 Jun-5-2002 Jun-4-2002 Jun-3-2002 May-31-2002 May-30-2002 May-29-2002 May-28-2002 Close 50.23 50.8 52.5 55.75 50.3 58.75 51.8 58.5 56.7 55.95 49.31 54.17 51.23 .1 48.9 51.0 51.55 53.91 56.47 52.3 50.63 50.

15 55.48 56.2 55.7 52.05 57.4 59.0 53.7 59.72 51.92 53.14 52.65 54.75 54.81 56.45 54.Sample Stock Market Data Date May-24-2002 May-23-2002 May-22-2002 May-21-2002 May-20-2002 May-17-2002 May-16-2002 May-15-2002 May-14-2002 May-13-2002 May-10-2002 May-9-2002 May-8-2002 May-7-2002 May-6-2002 May-3-2002 May-2-2002 May-1-2002 Apr-30-2002 Apr-29-2002 Apr-26-2002 Apr-25-2002 Apr-24-2002 Apr-23-2002 Apr-22-2002 Apr-19-2002 Apr-18-2002 Apr-17-2002 Apr-16-2002 Apr-15-2002 Apr-12-2002 Apr-11-2002 Apr-10-2002 Apr-9-2002 Close 59.65 58.93 53.56 58.8 53.48 55.0 54.87 53.98 53.0 54.31 .17 50.6 55.6 55.8 54.5 56.7 52.86 51.25 53.02 53.

n . Ccia3cdcdcga3g3a303g3a3cda3a3cdcdn3cc3a3a e . n. o. o r>.e I 1 10 lO lO CO 05 m e m n co m co in en N ^f OO 00 CO H O C 1 C O C < 0 ^ 0 ) 0 | 1 0 0 ' < * M N I » O cc 10 cM M M oM OO tP ^ C M i — csiC CM C M ^ H r H ^ H roHoO O C o oCc O C i^nl 'Ow iO l ' o m ' cNM M C O C c oCcO C O O O I ^ i M C n i n C C O c o ' O W i n i n t n m i n w w w m m m i n i n w m m m m m m i o m m L o m " 3 o o o o CM CM CM CM CM o o o o o o o o o " ~ o "o O o o C M C M C M C M C M C M C M C M C M C M C M C M C M C M ^ " ^ ^ ^ ^ O O N O ^ M H O O I O O U J ^ M CM CM CM CM CM CM CM oo m -^ co CM < > — < > — ' ' ' ' > ' ' < — ' > — ' > — ' v J C M C M C M C M C M C M C M C M C M C M C M C M C M C M CM CM CM CM CM i i i — i \ — i v — / o o o o o o O I OO N H ^ ^ ^ CO lO ^ iti o.

J^ * * >£. £. rf^ * £ Ol O l Cn ^1 CO CD to h-1 CO to cn co Cn Cn Cn -a H l rf^ 4^ ^r 00 bo o to 4^ *.* en CO *" *" co to ^ CO ^ Cn Cn Oi . rf^ J^ >t^ * * 00 ^J pj en ^ en * CO * en ^ * CO cn o CO en en ^ en CO bo 1 h^ 05 I— 02 to Cn to 1 — ** J £ rf^ ( ^ > .I oo to to to to to to to to to o o o o o o o o o o o o o o o o to to to to to to to to c^ p p p p to to to en CO to to to o o o o o o to to to f" * o* p p to oo to o o to p p to CD to o o to p p CO cp to o o to D p C p cr CO h-1 1 Fto to o o o o to to Oi ^ ^ * l > rf^ rf^ rf*.C-l C-l C-l C-l C H c-i C-l C-l t-i C-l C-l C-l C-l ou o p 9 p 5 P p ^ o o to 00 to o o1 h- 9 5 p co to h-» to to o o o o to h-' P P CO tO O o to c^ c^ 5-1 P P P 0 p ts p ~J oo 1 to to to o o o o O o to to to f * a P p CO to o o to c < Cw C_l 9 9 9 p p p 5 5 5 P P p p p p l—L 1—l 1 — » I—1 1—l 1—l H-* to l < cp 1— i Cn Oi .

86 49.3 40.36 45.55 46.200 An Undergraduate Introduction to Financial Mathematics Date Dec-27-2001 Dec-26-2001 Dec-24-2001 Dec-21-2001 Dec-20-2001 Dec-19-2001 Dec-18-2001 Dec-17-2001 Dec-14-2001 Dec-13-2001 Dec-12-2001 Dec-11-2001 Dec-10-2001 Dec-7-2001 Dec-6-2001 Dec-5-2001 Dec-4-2001 Dec-3-2001 Nov-30-2001 Nov-29-2001 Nov-28-2001 Nov-27-2001 Nov-26-2001 Nov-23-2001 Nov-21-2001 Nov-20-2001 Nov-19-2001 Nov-16-2001 Nov-15-2001 Nov-14-2001 Nov-13-2001 Nov-12-2001 Nov-9-2001 Nov-8-2001 Close 44.95 39.75 47.6 48.7 47.0 45.55 46.83 40.28 47.7 46.99 47.25 48.11 48.36 44.95 46.8 45.61 48.55 46.49 45.3 46.8 39.0 46.0 44.5 49.5 39.79 .2 45.14 46.2 45.11 44.5 46.

2 38.5 40.74 33.0 40.01 39.05 38.8 41.7 41.7 36.0 34.6 40.25 37.72 33.12 38.Sample Stock Market Data Date Nov-7-2001 Nov-6-2001 Nov-5-2001 Nov-2-2001 Nov-1-2001 Oct-31-2001 Oct-30-2001 Oct-29-2001 Oct-26-2001 Oct-25-2001 Oct-24-2001 Oct-23-2001 Oct-22-2001 Oct-19-2001 Oct-18-2001 Oct-17-2001 Oct-16-2001 Oct-15-2001 Oct-12-2001 Oct-11-2001 Oct-10-2001 Oct-9-2001 Oct-8-2001 Oct-5-2001 Oct-4-2001 Oct-3-2001 Oct-2-2001 Oct-1-2001 Sep-28-2001 Sep-27-2001 Sep-26-2001 Sep-25-2001 Sep-24-2001 Sep-21-2001 Close 39.43 40.19 36.75 .96 37.1 40.33 40.25 35.3 41.4 33.58 40.7 35.31 35.14 38.84 41.72 39.2 39.74 41.2 37.75 39.3 34.

56 37.48 42.48 51.An Undergraduate Introduction to Financial Mathematics Date Sep-20-2001 Sep-19-2001 Sep-18-2001 Sep-17-2001 Sep-10-2001 Sep-7-2001 Sep-6-2001 Sep-5-2001 Sep-4-2001 Aug-31-2001 Aug-30-2001 Aug-29-2001 Aug-28-2001 Aug-27-2001 Aug-24-2001 Aug-23-2001 Aug-22-2001 Aug-21-2001 Aug-20-2001 Aug-17-2001 Aug-16-2001 Aug-15-2001 Aug-14-2001 Aug-13-2001 Close 37.25 36.0 44.33 47.75 47.49 48.25 47.16 .1 48.0 40.92 47.91 46.2 42.9 50.9 45.7 44.4 41.35 37.55 47.72 47.1 49.65 50.

If there are n compounding periods per year.15. (2) The principal amount is P = $3993.93 (b) If the interest is compounded weekly n = 52 and A = 3993(1 + ^ .065. the annual interest rate is r = 0. the period is t = 2 years. where the final balance has been rounded to the nearest cent. Thus the final compound balance is P ( l + rf = 3659(1 + 0. (a) If the interest is compounded monthly n = 12 and A = 3993(1 + 2 ^ 5 ) 2 4 = $4350.l The Theory of Interest (1) The principal amount is P = $3659.Appendix B Solutions to Chapter Exercises B.57 (d) If the interest is compounded continuously A = Pert = 3993e(°-043^2) = $4351.^ ) 1 0 4 = $4351.065)' = $5013.60 203 .043.44 52 (c) If the interest is compounded daily n = 365 and A = 3993(1 + ^15)730 365 = $4351. the final compound balance is A = P(l + n -)nt. the period is t = 5 years. the annual interest rate is r = 0.

thus we wish to solve the following equation.004384268 = .08243216.204 An Undergraduate Introduction to Financial Mathematics (3) The period is t = 1. The effective simple annual interest rate R is 0. t « 42. (5) The continuously compounded interest-bearing account will increase in value more rapidly.058985832 = (1 + .^ ) 1 2 Vl.1 + (1 + —— ) 4 = 0.058985832 = 1 + — 1. The effective simple annual interest rate R is R = .0475t _ 1 0 0 0 ( 1 + ^0475)365t = : 365 This equation cannot be solved algebraically for t. 0. 365 Y00 = 0.0525.^ 0. the interest rate is r = 0.052611220 = i Thus if you pay an interest rate compounded monthly of at least 5.^ ) 1 2 1.058985832 = .6583 years. 1000eo. . (4) Since the competitor pays interest at rate r = 0.0525 compounded daily we will compute his effective simple annual interest rate and then determine the monthly compounded rate which matches it.261122% you will compete favorably.1 + (1 + .08 compounded n = 4 times per period. so we use Newton's method to approximate the solution.004384268 =1 + ^0.0538985832 365 R = -1 + 1+ Thus we wish to solve the following equation for i.

h—*0 (7) To determine the preferable investment we must determine the present value of the payouts.° n = 760. (8) If the price of the house is $200000 and your down payment is 20% then you will borrow P — $160000. Thus the exponent limit is 1 and lim(l + ft)1''1 = e 1 = e. ( ' 160000 = 1500— ( l r \ r w 0. h—>0 If we let y = (1 + h)l/h. Therefore lim(l + ft)1/. fixed rate mortgage should not exceed x = $1500 then we can substitute these quantities into Eq.8034% annually.12 Thus investment A is preferable.055 + lOSe" 0 0 8 2 5 + 2 0 5 e . then lny = l n ( ( l + ft)1/'') = i ln(l + ft) ny _= e *ln(l+h) l n ( l + h) y = e = h . (1.108034 i + n r n-(12)(30) 1 + l2 Thus the interest rate must not exceed 10.° n = 760.Solutions to Chapter Exercises 205 (6) We focus on finding the limit lim (1 + h)l/h.° 0 8 2 5 + 2 0 0 e . If the monthly payment on a t = 30 year. . Since the interest rate is r = 0.6) and use Newton's Method to approximate r.25 PB = 198e-° 0 2 7 5 + 205e-° 0 5 5 + 2 1 1 e . T-[ —nt ' = * . The limit in the exponent is the definition of the derivative of In x at x = 1.l = l i m e " T /I->0 ln(l + fe) h^O = e since the exponential function is continuous everywhere.0275 compounded continuously we have PA = 200e-° 0 2 7 5 + 211e-°.

7) we setup the equation 2000 3000 4000 2 3000 3 1+ r (1 + r) (1+r) (1 + r) 4 whose solution is approximated using Newton's Method. M~i)(l i=l + r). Thus if there exist two distinct rates of return n and ri then by the Mean Value Theorem / ' ( r ) = 0 for some r between T\ and V2. (1. B. If you are concerned about the problem of keeping track of which is the first die and which is the second. The rate of return r ss 0. (b) The rate of return of investment A is TA = 0. According to the Multiplication Rule the probability of the event (m. —$10.' 11 =-J2 i=l Mm + r)~iz+1) <0 for — 1 < r < co.206 An Undergraduate Introduction to Financial Mathematics (9) Differentiating the function / ( r ) given in Eq.049921.0433).0717859 or equivalently 7. The present value of the initial investment and payout amounts for investment B is $18. Thus investment B is preferable if present value is the decision criterion.91.63.n) represent the numeric outcomes of the rolling of the dice. n) is P (m. (2) Let the ordered pair (m.039354 while the rate of return of investment B is TB = 0. (11) (a) The present value of initial investment and payout amounts for investment A is (assuming a simple annual interest rate of r = 0. The outcome of the first die is m and the outcome of the second die is n. the probability of it landing on 3 is p = 1/4. contradicting the inequality above.17859% per year. assume that the first die is painted green while the second is painted red. (10) Using Eq. Thus investment B would be the preferable investment if rate of return were the deciding criterion. (1. n) = P (first die shows m) P (second die shows n) .7) yields n n /'(r) = Y.2 Discrete Probability (1) Assuming the regular tetrahedron is fair so that it equally likely to land on any of its four faces.

2) (3. Each of the entries in the previous table occurs with probability 1/36 and therefore the probabilities of the outcomes of rolling a pair of fair dice are Outcome 2 3 4 5 6 Prob. 9.3) (1.6) (2.5) (3.5) (6. n) = 1/36.4) (1.5) (5.4) (6.6) (6. 4.4) (5.1) (5.1) (3.3) (4.5) (1.1) (4.1) (1.Solutions to Chapter Exercises 207 since the dice are independent of each other. 4 5 6 7 8 9 5 6 7 8 9 10 6 7 8 9 10 11 7 8 9 10 11 12 Thus we can see the sample space of sums for the fair dice is the set {2. 3.3) (2.3) (5.4) (4.1) (2. 7. The table of 36 outcomes is shown below. l 36 1 18 1 12 1 9 5 36 Outcome 7 8 9 10 11 12 Prob. 8.2) (6.6) Since we are interested in the sums of the numbers shown on the upward faces of the dice we will tabulate them below.4) (2.5) (4.5) (2.4) (3.3) (6.2) (5. then P (TO.6) (3.2) (4.12}.6) (4. Let the probability that the batter strikes out .2) (2.10.11. 5. 6. (1. Since each of the die has six equally likely simple outcomes. l 6 5 36 1 9 1 12 1 18 1 36 (3) Let the probability that the batter strikes out in the first inning be denoted P (1) = 1/3.2) (1.1) (6.6) (5.3) (3.

52 J 1 ~ 221' . Thus the probability of winning under the strategy is 6/8 = 3/4 = 0.75. Thus one of the three will guess the correct color in these six cases. According to the Addition Rule the probability that the batter strikes out in either the first or the fifth inning is P(lV5)=P(l) + P(5)-P(lA5)=i + i . Person 1 Red Red Red Blue Red Blue Blue Blue 2 Red Red Blue Red Blue Blue Red Blue 3 Red Blue Red Red Blue Red Blue Blue Hat In 6 of the outcomes two of the three people will see their companions wearing mis-matched hats and pass while the third will see their two companions wearing matching hats opposite in color to their own hat's color. Let the probability that the batter strikes out in both innings be denoted P (1 A 5) = 1/10. (5) Since the cards are drawn without replacement the outcome of the second draw is dependent on the outcome of the first draw.^ = | . In the remaining two cases all three hats are of the same color. Using the Multiplication Rule and conditional probability we have P (2 aces) = P (2nd acejlst ace) P (1st ace) = [bl) (.208 An Undergraduate Introduction to Financial Mathematics in the fifth inning be denoted P (5) = 1/4. so the strategy of guessing the opposite color would fail. (4) The eight possible outcomes of the experiment are listed in the table below.

P (1) = 1/13. Using the Multiplication Rule and conditional probability we have P (2nd ace A 1st not ace) = P (2nd ace 11st not ace) P (1st not ace) 4 \ /48 N 51 J V 52 16 221' (7) Since the cards are drawn without replacement the outcome of the fourth draw is dependent on the outcome of the third draw which is dependent on the outcome of the second draw which is dependent on the outcome of the first draw.51J \52j _ 1 ~ 270725' (8) If n £ { 2 .50 J (. . P (49) we obtain the following probabilities for the first ace appearing on the nth draw. . Using the Multiplication Rule and conditional probability we have P (4 aces) = P (4th ace 13rd ace) • P (3rd ace 12nd ace) • P (2nd ace11st ace) • P (1st ace) = (49J (. . 49} is the position of the first ace drawn then 48 • 47 • 46 • • • (50 . . . . . . .Solutions to Chapter Exercises 209 (6) Since the cards are drawn without replacement the outcome of the second draw is dependent on the outcome of the first draw.n) • 4 52-51-50---(53-n) (n) When n = 1. P ( 3 ) . By calculating P (2).

2520 thus c = 2520/7381.210 An Undergraduate Introduction to Financial Mathematics P(n) 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 16 221 376 5525 17296 270725 3243 54145 3036 54145 2838 54145 1892 38675 1763 38675 328 7735 164 4165 152 4165 703 20825 8436 270725 222 7735 12 455 11 455 352 15925 5456 270725 992 54145 899 54145 116 7735 522 38675 36 2975 9 P(n) 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 833 92 10829 2024 270725 253 38675 44 7735 38 7735 228 54145 57 15925 15925 48 8 3185 16 7735 1 595 4 2975 22 20825 44 54145 33 54145 24 54145 12 38675 8 38675 1 7735 4 54145 2 54145 4 of being the 270725 1 270725 The draw with the highest probability first ace is the first draw. thus the probability that any spin of the wheel has an outcome of black is P (black) = 9/19. (9) The outcomes of different spins of a roulette wheel are assumed to be independent. (11) To be a probability distribution function 10 10 10 2=1 7381c = 1. . (10) The outcomes of different spins of a roulette wheel are assumed to be independent. thus the probability that any spin of the wheel has an outcome of 00 is P (00) = 1/38.

49 n—1 5 3 H = J2 (n • P (n)) = — = 10.0.000749405.« 2. Thus the expected number of female children in a family of 6 total children is 120 = —.Solutions to Chapter Exercises 211 (12) P (0 black) P (1 black) P (2 black) P (3 black) 5 4 3 1 20 ' 19 18 ~ 118 ' 15 5 4 5 15 4 5 4 15 5 : 20 '19' 18 + 20 ' 19''18+ 20 19 ' 18 ~ 38 15 14 5 15 5 14 5 15 14 35 ' 20 '19 '18 + 20 ' 19 ' 18 + 20 19 18 ~76 ' ' 15 14 13 91 20 ' 19 18 ~ 228 ' (13) This situation can be thought of as a binomial experiment with 25 trials and the probability of success on a single trial of 0.92683. (14) The probability of a child being born female is p = 100/205 = 20/41.0016)25"71 w 0.6. 41 (15) Using the probabilities calculated in exercise 8 and the definition of expected value we have the expected position of the first ace. Thus the probability that a manufacturing run will be rejected is 25 P {X > 2) = J^ n=2 25 P (X = Yl ( ^ ) (°.0016. .0016 ) n ( 1 .

b2 = a2E [X2] . 2 2 TW ^ 2 901 2809 1696 „„„„ ^(n .3 Normal Random Variables and Probability (1) The probability distribution function for X is given by the piecewise defined function /(X) = .a 2 (E [X])2 = a 2 Var (X) B. n=\ This implies the standard deviation in the appearance of the first ace is approximately 8.2365.2a6E [X] .P(n))-M2 = — .212 An Undergraduate Introduction to Financial Mathematics (16) By the definition of expected value E [aX + b] = J2 ((aX + b)F (x)) x = £(aXP(X)+6PpO) x = 5>XP(X))+^>P(X)) x x = a£(XPPO)+&X>PO x = aE [X] + b x (17) Using the probabilities calculated in exercise 8 and the expected value calculated in exercise 15. .(aE [X] + bf = a 2 E [X2] + 2abE [X] + b2 -a2 (E [X])2 . (18) By the definition of variance Var (aX + b) = E [(aX + b)2] . the variance in the position of the first ace drawn is s^ .— = —=67.84.H-4<*<I I 0 otherwise.(E [aX + b})2 = E [a2X2 + 2abX + b2] .

/ -OO />0O f(x)dx+ f{x) (£E = P (X < a) + P (X > a) 1 . (3) OO /(a:) da.2)' / ( * ) • Show that the mean of this random variable does not exist.Solutions to Chapter Exercises 213 Therefore . thus J(x)dx = J^ •-$ dx c lim MM> dx —- c lim = C ivlToo(2ii + ^ 2 c = 2.P (X < a) = P (X > a) (4) A random variable X has a continuous Cauchy distribution with probability density function 1 7r(l + a.00 Z"1 1 1 P(X>0)=/ Jo f(x)dx = -dx = -. . Jo 5 5 (2) We know that for any random variable X with probability distribution function f(x) we must have f_ f(x) dx = 1.

/ = a / J—OO (bf(x))dx f(x) dx (xf(x)) dx + b J = aE [X] + 6 . (5) E [aX + b] — OO OO (ax -h b)f(x) dx (axf(x) + bf(x)) dx /•OO (axf (x)) dx-\. the mean of a continuously distributed random variable is oo xf(x)dx.2) dx + J_ocn(l+x2)dX M ^ .o o JM J0 TT(1 + x2) N->ooJ0 dx dx N\ N lim lim 7r(l + X2) dx + lim 7r(l + X2) \n(i+xi) M . / -oo Thus for the given probability distribution function r(l + a.214 An Undergraduate Introduction to Financial Mathematics By definition. M—> —oo \ Z7T + lim — ln(l + x2) lim [ — ln(l + M 2 ) ) + lim ( — ln(l + 7V2' The improper integral above does not converge and hence the mean of the random variable does not exist.

.. . .. -X be pairwise independent. .x2.. . . n (7) Proof. where 00 />oo />oo fi(xi) = I / OO -/ —OO •••/ •/ —OO f(x1.xk)dxi .. X2..Xfc)^xi dr 2 • • .dxk x2f2(x2)dx2-\ h / xkfk(xk)dxk xifi(x1)dxi+ ) J — OO = E [Xx] + E [X2] + • • • + E [Xfe] . . Since the component random variables are assumed to be pairwise independent the joint probability distribution can be rewritten as f{x1.... -dxk x i / ( x i .x2. To keep the notation compact we have written the marginal probability density of Xi as fi(xi) for i = 1..Solutions to Chapter Exercises 215 (6) Proof.. •••. Xk)... By the definition of expected value E[X1+X2 OO + ---Xk] /"CO (xi +x2 H -00</—OO OO /"OO •/ — OO />0O \.... X2.. Let X L ...x2. . Let X\. continuous random variables with probability distribution function f(xi..x2..xk) = fi(xi)f2(x2)---fk(xk)..da?fc 00 ^ — 00 OO /"OO />00 / / -co J — 00 00 *••/ J — 00 />oo 0 0 1/ — 0 0 x2f(xi. .Xk) dxi dx2 .xk)dxidx2 ...xk)dxi dx2 .2..x2..Xk)f(xi. • • • . dx* xkf(x1.x2.dxi-idxi+i ... .k.... -. x 2 ....xk).dxk.x2. Xk be continuous random variables with probability distribution function f{x\. .

0 0 «/ — 0 0 •••/ J — 00 x1fi(x1)x2f2(x2)---xkfk(xk)dxidx2...x2. Let X be a continuous random variable with probability distribution function / ( x ) ..dxk x2f2(x2) dx2 J ••• f / xkfk{xk)dxk D = ( / a.) da. = E [X 2 ] .2/x2 + = E [X 2 ] M2 A*2 • .M ) 2 /(20efe (x2 -2nx + /j2)f(x)dx /"OO /'OO -CO / O O -CO CO / -CO a: 2 /(x) dx-2fi J— OO xf{x) dx +/J2 / J —OO /(a...xk)dxidx2.. (8) Proof..i/i(xi)da.dxk / OO / -OO J —OO /"OO •••/ J — OO /"OO xix2-••xkh(x1)f2(x2)---fk{xk)dxidx2.-dxk / / . By the definition of variance oo / (:c.. Suppose the E [X] = fi.i J ( / = E[X1]E[X2]---E[Xk].216 An Undergraduate Introduction to Financial Mathematics Thus by the definition of expected value E [ X 1 X 2 • • • Xk] oo />oo /-oo / / -00 J — 00 CO /"CO •••/ </—00 /*CO x1x2---xkf(x1..

(a2 (E [X])2 + 2abE [X] + b2) oo />oo />oo / -oo x 2 /(x) dx + 2ab / J — oo a / ( a ) dx + b2 / J— oo / ( x ) dx -a2(E[X])2 -2abE[X] . By .a 2 (E [X])2 .2. fc. Var (aX + b)=E [(aX + b)2} . .. be /ij for i = 1. . X2. continuous random variables with probability distribution function f(x\. Let X i . Xk be pairwise independent..(aE [X] + b)2 (a 2 x 2 + 2abx + b2)f{x) dx / -OO / O O -OO . Suppose a... 6 € R. Let the mean of X.b2 = a2(E[X2]-(E[X])2) = a 2 Var (X) D (10) Proof. .Solutions to Chapter Exercises 217 (9) Proof.. Let X be a continuous random variable with probability distribution function f(x).(E [aX + b]f OO {ax + b)2f(x) dx ... ..X2.2abE [X] . Xfc).b2 = a 2 E [X 2 ] + 2a6E [X] + 62 .

.(E [Xi + X 2 + • • • + Xfc])2 oo /*oo / -•..(i4 + /f| H .. + 2 2 1.+ /xfe)2 / •••/ -co CO (xi H /*00 + Xk)2fi{xi) •' • fk(%k) dxi.(/xi + • • • + /Xfc)2 / / -CO J —OO •••/ J —OO (a.. + 2/xi/x2 H 2 2 = E [Xi2] + E [X22] + • • • + E [Xl] + 2/xi/X2 + • • • + 2/ifc_i/ifc 2/Xfc_i/xfe) 2/ifc_i/i fe ) = E[X ]-/x + E[X ]-/x + ---+E[X2]-/x2 = Var (Xi) + Var (X2) + • • • + Var (Xfe) D .+ a .+ E[X 2 ] + 2E [XiX2] + • • • + 2E [Xfc_iXfc] (/x2 + n\ H 2 2 1./ 4 + 2/xi/X2 H H /x2.. dxk J —oo /-CO .dxk h / 4 + 2/xi/t 2 H 2/jfc_i/Xfc) ./x2. | + 2x112 H 2 h 2xk-iXk)f\{xi)f2{x2) • • • fk(xk) dxi dx2-.(/J + nl H = E [ X 2 ] + E [ X 2 2 ] + ..(/x2 + /z?... + 2/X 1 /J 2 H 2 2/x/c_i/Xfc) = E [ X i ] + E [ X ] + . + E[X ] + 2E [Xi] E [X2] + • • • + 2E [Xfc_x] E [Xfc] ...f + a..xk)dx1 .| + --./ -co CO (xi H \-xk)2f(x1..218 An Undergraduate Introduction to Financial Mathematics the definition of variance Var (Xi + X2 + • • • + Xk) = E (Xi + X2 + • • • + Xfc)2] . .dxk J — oo /-CO -(/xi + --.

ii2 .) 196 225 196 225 2 f / 5 (-x3)dx+ x* "T 10 17 196 ~ 10 ~ 225 ~ _ 373 ~ 450' + 225 = 1 (V1 + 1 . -x 2 ) da.x 2 | a .^ (12) Proof. This is equivalent to the supposition that n — m = 2k for some fceZ.Solutions to Chapter Exercises 219 (11) The expected value of X is calculated as oo xf{x) dx / -OO r 2 xlxl da. n — m = 2k n — m + 2m = 2k + 2m n + m = 2(k + m) . Suppose ra.3 da.m i s even.1 + 8) M 15 14 15 oo The variance is calculated as / x2 f(x) dx . ar 2 ' '° 5 l jf" da. | dx • l 5 -OO L 196 225 / a. 6 ) .neZ and n . !i _2_ "3 ( .

e 4fct dx = kt J 2Vknt J /knt :e-udu du ?/- e~u + C where C is a constant of integration. Using the technique of integration by parts with X . we obtain v = —kte x i 2 dx dv = — e~"« dx . where u = x /{Akt) we obtain / —. 2 TO) • (13) Using the method of integration by substitution. that n +TO= 2j for some j € Z. (14) Use the technique of integration by parts to evaluate -e 4ht / 2Vkirt dx. Now suppose that n +TOis even. or equivalently.220 An Undergraduate Introduction to Financial Mathematics Thus n + 77i is even. _j£i lkt vknt 1 du = . n + m = 2j n + m — 2m = 2j — 2m n — m = 2 (J — Thus n —TOis even.

157305.. (16) The probability distribution function for the standard normal random variable is and thus 1 (a) P ( .. (d) P (1 < X < 3) = .9973.= / e~x^2 dx « 0..682689. V27T J _ i (b) P ( ..7 = / 1 /""^ e-^ 2 / 2 dx « 0.9545. lim 2kt M e M2/4fct 2/ct M—KX MeM lim ^ 2 /4kt = 0. V 27T J l (17) Proof.= 1 f1 / Z" 2 e~x2/2 dx « 0. .J —oo r>oo / we u2 du / -oo ue u2 "2" tiu 0.. 2ktM lim .Solutions to Chapter Exercises 221 (15) Assuming that k > 0 and t > 0 we have M->oo lim 2ktMe~M '4kt = lim 2ktM M2 ikt e M^oo / ' The limit on the right-hand side of the equation is indeterminate of the form oo/oo and thus we apply L'Hopital's Rule. Suppose X is a normally distributed random variable with mean \i and variance a2.K X < 1) = . M ^ o o e M 2 /4fct = M-*oo . E[Z}=E[g(X)} l r° aV2~n J-( 1 1 X — fl a /•oo e _{x-lx)-1 2<T^ dx Making the substitution u = (x — fi)/a yields E[Z] = = aV2n ..2 < X < 2) = — = / e" x 2 / 2 da...3 < X < 3) = . V27r 7-2 1 f3 (c) P ( . We will let z = g{x) = (x — [i)jo. « 0.

assuming that rainfalls in different years are independent.44192) = P (Z > 0. Likewise a2 = Var (X) = Var (Xi)+Var (X2) = (J\+<J\ = (3.2) 2 +(3. = E [X] = E [Xi] + E [X2] = Mi + M2 = 28.2) 2 = 20. then JJ.44192) . Let the random variable X be the sum of the rainfalls in two consecutive years.44192) = 1-P{Z « 0. The rainfalls are normally distributed random variables.2.222 An Undergraduate Introduction to Financial Mathematics Therefore Var(Z) = E [ Z 2 ] -{E[Z})2 = E [(g(X))2] aV2n i-oo V °" / D (18) Think of the consecutive years as year 1 and 2. call them X\ and X2 in each year with means £ti = /Z2 = 14 and standard deviations o\ = 01 = 3. Now let Z = (X — fi)/a and then P(Z>30) = / P ( Z 30-28\ > 7 = ) < 0.48.0(0.329266 = 1 .

Solutions to Chapter Exercises 223 (19) Let the random variable X represent the ratio of consecutive days selling prices of the security. (20) Since X is uniformly distributed on [a. The probability of a four-day decrease in the selling price of the security is P(X4 < 1) = P ( 4 1 n X < 0 ) p|Z<°-a04 0.57926.9.05 = P(Z>--0.42074.57926 Thus the probability of a one-day increase in the price of the security is approximately 0.20) « 0. b] its probability distribution function is /•/ \ f TT— H a < x < b.40) 0. f J X 1 if x < a if x > b fit) ={ dt |Ef if a < ^ < 6 0 . f(X) = ) 0 b~a otherwiseWe will make use of Theorem 3.10 P {Z < -0.344578. Consequently the probability of a one-day decrease in the price of the security is approximately 1 — 0.01 0. p {X > 1) = P ( l n X > 0 ) = = p(z> p(z> *-¥) 0-0.57926 = 0.

Using the definition of variance and Eqs. ) V 2 ^ 2TT Br^'^-M^))' D .( E [ ( X . ( M .17) we have Var ({X . 1 rX e-*'/*dt= 1 2TT.2K)2 + a2^ .K 2(b-a) ifa<K<b if K > b UK <a if a < K <b if K > b 0 (21) By definition -''Vdt.224 An Undergraduate Introduction to Financial Mathematics Thus E [(X . V27T j .X ) + ] ) ((/* .K)+] = j°° (J°° f(t) dt\ dx JbK^adx 0 ^ .2 X ) a p _ ( M _ 2 J . (3.{V-2K\ A a.K)+) 2 2 E ((X-i^)+) .15) and (3.4>{x) (22) Proof. ~u2/2du 2nJx -u2/2 V2TT du 4>{-x) = 1 .o o Now we make the substitution u = —t.

2Ke>M+a2/2(f)(w + a) + K2(j>(w) /eM+a2/20 (w + a)- K<j) {w)Y where w = [ji — h\K)/a. (3.19) we have Var ({X . y = 1. • B.4 The Arbitrage Theorem (1) Converting from the odds against to probabilities that the outcomes may occur produces a table of probabilities like the following: Outcome A B C Probability Since the probabilities sum to ^§ > 1 the Arbitrage Theorem guarantees the existence of a betting strategy which yields a positive payoff regardless of the outcome. Winning Outcome Payoff -2x + y + z x — 2>y + z x+ y — z A B C We need only find values for x.Solutions to Chapter Exercises 225 (23) Proof.16) and (3. and z which make the three expressions in the column on the right positive. y. The payoffs under different winning scenarios are listed in the next table. and z = 2 is one solution to this set of inequalities. and C respectively. y. .K)+) E ((X-K)+f -(E[(X~K)+}Y = e2(M+<j2)0(™ + 2a) . Using the definition of variance and Eqs. B. and z on outcomes A. By inspection we can see that x = 1. Suppose we wager x.

There are three corners to the region located at (0.0).6). x2 10 10 xl .3/2). (3/2.226 An Undergraduate Introduction to Financial Mathematics (2) The region in convex and unbounded. and (6. The region is sketched below.

This occurs when (xi.3/2). x2 10.Solutions to Chapter Exercises 227 (3) The cost function will be minimized at the smallest value of k for which the graph of x\ +x2 = k intersects the feasible region sketched in exercise 2. —^ 10 xl . x2) = (3/2. Thus the minimum cost is k = 3.

5 2.5 1.228 An Undergraduate Introduction to Financial Mathematics (4) 0 < xi < 2 and 0 < x2 < 3 x2 4r -0.5 0.5 xl -1L (5) We can think of the solution as a system of inequalities: xi > 0 x2 > 0 x3 > 0 xA > 0 x 2 + £4 = 3 where x3 and £4 are the newly introduced slack variables. The last two equations can be re-written in matrix/vector form as ' Xl' Ax "10 10" 0 1 0 1_ x2 £3 _X\_ = "2" 3 .

X4) subject to 1 0 02 X\ X4 5 1 T h u s the minimum value of the cost function is 0 and occurs where (zi.2. T h u s we want to think of the primal problem as: Minimize x\JrX2~Vxz subject to 2x\+X2 with xt > 0 for i = 1. 0) • (xi. Therefore the primal problem can be restated as the following: Minimize: (0.2 x2 = x-$ = 0.XA). From the inequality constraint in the dual problem we know t h a t y < 1/2.x3). can be thought of as the primal problem: Minimize: c • x T h e problem as stated subject to A x = [ 1 2 3] x = 15 = 6.X2. This is equivalent to the dual problem: Maximize: by = 15y subject to yA < c.X3.x2. (8) In this case we must introduce a slack variable X4 to convert the inequality constraint in the primal problem to an equality constraint.4. T h e problem as stated can be thought of as the primal problem: Minimize: c • x subject to Ax = "1110" |_0 1 1 2\ x = ll\ "5" This is equivalent to the dual problem: Maximize: y • b = [yi y2) b subject to yA < c.2) and x = (xi. = 4 and £3 + 2:4 = 6 .1. (7) Let vector c = ( 0 .0.Solutions to Chapter Exercises 229 (6) Let vector c = (1.X2. T h u s the m a x i m u m value of 15y is 15/2. 0 ) and x = {XI.3. T h u s by Theorem 4.1/2). T h e constraint inequality for the dual problem is equivalent to the following system of linear inequalities.x3. 9 .0. 7 . 2/i < 0 2/i + 2/2 < 7 2/1 + 2/2 < 9 2|/ 2 <0 T h e first and last of these inequalities imply strict inequality in the second and third. T h u s the minimum value of c • x = 15/2.a:4) = (5.

230 An Undergraduate Introduction to Financial Mathematics If we define the following vectors and matrix (1. _4_ .0).2/3) • (2.£3.6).This is equivalent to maximizing y1 + 62/2 subject to the system of inequalities 2yi < 1 2/i < 1 2/2 < 1 2/2 < 0 . ° 2.2:2. We can see from the constraint equation for the primal problem that x = (2:1. (10) Written in matrix/vector form this dual problem becomes: Maximize y • b = (2/1.1. .. (11) Find the solutions to the primal and dual problems of exercise 10.£4) (2. Thus the primal problem is Minimize c • x = (1. then the primal problem is the one of minimizing c • x subject to Ax. A 2 100 00 11 and b = (1. 0.0) • (£1.X2. Thus by Theorem 4.x4}.-. 6) and the minimum value of the cost function is 2. !)• Thus the minimum value of the cost function is .2 X2 = 2:3 = 0. The dual to this problem is the one of maximizing y • b subject to yA < c where y = (2/1. 4) subject to " 1 0" ' 1" yA = [2/1 2/2 2/3. We can also see that the maximum of y • b will be 1/2.X2.1. = b and x > 0. X2) = (1. —1) • {x\. x = (x1.1 1 < .x2) ' 1 0" Ax. However. (9) Paying attention to the inequality constraints of the dual problem indicates that 2/1 < 1/2 and 2/2 < 0.0. subject to "2" = 0 = b. we can also solve the primal problem directly after re-writing it as Minimize: (1. ° 2.1 1 Xi .2/2. This is also the minimum of the primal problem.0:4) subject to "20' 01 Xi X4 = "4" 6 From the constraint equation we see that now (2:1.0.x3.2/2).1 .

67. the desired probability is equivalent to determining the probability that a symmetric random walk initially in state 5(0) = 50 will attain a value of 75 while avoiding the absorbing boundary at 0.Solutions to Chapter Exercises 231 c • x = (1. 75] which begins in the state 5(0) = 50. the probability is p 75 (50) = 50/75 = 2/3. (5) As in exercise 3 the conditional exit time is equivalent to the conditional exit time of a symmetric random walk on the discrete interval [0. (5. b. then g must be a quadratic function of the form g(x) — ax2 + bx + c where a. Using Theorem 5. and c are constants.75] which begins in the state 5(0) = 50. the probability is fi0. Since f(A) = 1 then a = 1/A Thus we have f(x) = x/A. Since g(A) = 0 then b = A. A]. Thus Q7s(50) = 3125/3 « 1041. Therefore the maximum of the dual problem is likewise 0. Because of the spatial homogeneity property of the random walk. twice differentiable function whose second derivative is —2 on the interval [0. symmetric random walk. B. (3) This is an example of a symmetric random walk. Thus we have g(x) = x(A — x). This is best done with the aid of a computer or calculator since the linear system contains 76 unknowns (really only 74 since the boundary conditions are known). We must set up and solve a tridiagonal linear system of the form shown in Eq. In this case Z50 = 6250/9 « 694.1. A]. Since /(0) = 0 then 6 = 0. This agrees with our earlier derivation of the conditional stopping times for the discrete. — 1) • (1. (4) As in exercise 3 the exit time is equivalent to the exit time of a symmetric random walk on the discrete interval [0.5 Random Walks and Brownian Motion (1) If / is a continuous. Using Theorem 5. then / must be a linear function of the form f(x) = ax + b where a and b are constants.2. This agrees with our earlier derivation of exit probabilities for the discrete.50) = 1250. Since g"(x) = —2 then a = — 1 which implies g(x) = x(b — x).1) = 0. (2) If g is a continuous. twice differentiable function whose second derivative is 0 on the interval [0.15) where A = 75. Since g(0) = 0 then c = 0. symmetric random walk.444.75(50) = 50(75 . .

A histogram of AX = lnP(ij+i) — lnP(ij) is shown below.0332183 day" 1 respectively. The random walk generated is pictured below.1 respectively. The mean and standard deviation of AX are fiAX = 0.yahoo.028139 d a y . (8) From the website finance.00121091 day" 1 and aAX = 0. Using the data collected in Appendix A.25) and set P(to) = 42. the closing stock price on the last day for which data was collected was $42. (7) Various results are possible due to the fact that we are simulating a stochastic process. . If we substitute these values in Eq. Inc.com the closing prices of stock for Continental Airlines. (symbol CAL) were captured for the time period of 06/15/2004 until 06/15/2005.86.232 An Undergraduate Introduction to Financial Mathematics (6) Let P = P(0)e"<.000555 d a y .1 and a = 0. From the data the estimated drift parameter and volatility are \x = -0. (5.86 with At = 1 we can iteratively generate 252 (or any other amount for that matter) new days' closing prices. then dP_ dt d ~d~t (P(OK') = P(0)e"t|(Mi) = P(0)eMV = tiP.

1 -0. The random walk generated is pictured below. (5.0332183 day~* respectively. The closing stock price on the last day for which data was collected was $10. If we substitute these values in Eq.00121091 day" 1 and a = 0. From the data the estimated drift parameter and volatility are \i = 0.21 with At = 1 we can iteratively generate 252 (or any other amount for that matter) new days' closing prices.Solutions to Chapter Exercises 233 60 50 40 • 30 20 10 l 1 .1 (9) Various results are possible due to the fact that we are simulating a stochastic process. . 250 .21.25) and set P(£ 0 ) = 10.

z of + By definitionFyy{y0. z0 + xk)— [z0 + xk] ax = Fy(y0 + xh.z0 + xk). z0 + xk) — [y0 + xh] + Fz(y0 + xh. z0 + xk) — = Fy(y0 + xh. z0 + xk)— [z0 + xk] + kFzy(y0 + xh. z0 + xk)h + Fz(y0 + xh.z0) + remainder 0) order k Fzz(y0. z) and f(x) = F(y0 + xh. z0 + xk) + xh. 3 + xh. z0)k.z0) is Rs(x) — 3 for f(x) -x for some a between 0 and x. z0 + xk) + 2hkFyz(y0 + k Fzz(y0 2 „ / " ' ( « )) . z0 + xk) + 3h2kFyyz(y0 + 3hk Fyzz(y0 2 + xh.234 An Undergraduate Introduction to Financial Mathematics (10) If F = F{y. If we notice the binomial 3! coefficient pattern among the partial derivatives in the first and second derivatives of f(x) then we can readily see that f'"(x) = h3Fyyy(yo + xh. z0 + xk)— + Fz(yQ + xh. z0 + xk) then f'(x) dv dz = Fy(y0 + xh. z0 + xk)— [y0 + xh) + kFzz (y0 + xh. z0 + xk)k—~ = hFyy(y0 + xh. Differentiating once again produces dij dz f"{x) = Fyy(y0 + xh. . z0 + xk) 2 /"(0) = h the Taylor 2hkFyz(y0. z0 +xk)k— + Fzz(y0 + xh. z0 + xk)h— + Fyz(y0 + xh. z0)h + Fz{y0. z0 + xk)k /'(0) = Fy(y0.z0 + xk) + k Fzzz(y0 3 From this expression we may directly determine the Taylor remainder of order 3. z0 + xk) 2 + xh. z0 + xk) — [z0 + xk] ax = h2Fyy(yo + xh. z0 + xk)— [y0 + xh] + hFyz(y0 + xh. z0 + xk)h— dij dz + Fzy(y0 + xh. + xh.

t) = fiP and b(P. An investor could sell the European put option and buy the American put option and invest the net positive cash flow of Pe — Pa in a risk-free bond at interest rate r. At the exercise time.Solutions to Chapter Exercises 235 (11) If we apply Ito's Lemma with a(P. in either the case where the owner of the European put exercises the option or not. then dY = (nPnP™-1 + -(aP)2n{n nfJLpn + n n . . (2) The net payoff will be the quantity (S(T) . the investor has a portfolio with value (Pe .t) = aP and Y = Pn.6 Options (1) Suppose the value of an American put option is Pa and that Pa < Pe. the value of a European put option on the same underlying security with identical strike price and exercise time.K)+ .t) = /iP and b(P. The payoff is plotted below.t) = aP and Y = In P.l)Pn~A dt + n(Jpn dt + aPnPn~x d w dW(t) ( ~ 1> pn\ ^ (12) If we apply Ito's Lemma with a(P.C where K = 60 and C = 10.Pa)erT > 0. then dY fiP "ln ~PJ + \(<rPf )dt + aP "1 dW(t) M+ a2N ] dt + a dWit) B.

the strike price be K.236 An Undergraduate Introduction to Financial Mathematics payoff 30f 20 10 S(T) 20 •10 40 60 80 100 (3) Suppose C is the value of a call option on a security whose value is S. (4) Since Ce > S .S'(0))ert* .06 » 0 . Suppose at some time. then the investor sells the security for min{5(i*). If the owner of the call option chooses to exercise the option at time 0 < t* < T. Hence at the time of exercise the investor's portfolio has a value of + (C . say t = 0. r = 0. Let the strike time of the option be T. then an investor could take a long position in the the security and a short position in the call option.26e-(°. K}.Ke~rT T = 1/4.25 ) > 3.06. and the risk-free interest rate be r. K = 26. then when S = 29.38709 . and Ce > 29 . This would generate cash for the investor in the amount of C — 5(0) > 0 which could be invested in a risk-free bond at interest rate r.min{S{t*). it is the case that C > S. Thus at time t the investor owns a portfolio worth S(t) + (C — S(0))ert.K} S(f) > 0.

(6.25 ) K = 31.07 then according to the Put-Call Parity Formula in Eq.25) + 3 0 _ 30e(o.S = 30 + 3 .10 then according to the Put-Call Parity Formula in Eq.io)(o. This amount could be invested at the risk-free rate until the strike time arrives. This generates an initial cash flow of K + C e .io)(o. Ce = 3. if = 14.83762 (8) An investor could take a long position in the put option and the security and short position in the call option while borrowing.1): Pe = Ce-S = + Ke~rT + > -S + Ke~rT _H l4e-(0-07)/6 = 2. C e = 3.10 )(°.25 + 31 = 3 + ^e-(0-io)(o. S = 11. K = 31.2 5 )-31 Pe = 2.Solutions to Chapter Exercises 237 (5) Suppose S = 31. and r = 0. .io)(o.1 . at the risk-free rate. T = 1/4.25) P e = 3 + 31 e -(°.0158 (7) Suppose T = 1/6.25. T = 1/4. Pe = 2.io)(o.10 then according to the Put-Call Parity Formula in Eq. At t = 3 months if S(3) > 30 = K. (6.25) 2.Pe .31 = $1.25 + 31 .23461 (6) Suppose S = 31.1 0 » 0 . 2 6 5 8 6 2 > 0. (6.3 = Ke-(o.1): Pe + S = Ce + Ke~rT 2.25) = $ 0 .25) K = (2.25 + 31-3) e (°. and r = 0. and r = 0. the strike price of the options. then the call option will be exercised at the investor cancels all positions with a portfolio worth le(o.1): Pe + S = Ce + Ke~rT P e + 31 = 3 + 31e-(o.

and rfi = h.s- and Adding left-hand sides and right-hand sides of the previous two equations produces rcifi + rc2f2 = ci/i.e. (9) If fi(S. As the price of the stock increases toward infinity the put option will not be exercised and thus linis^oo F(S.ss) +c2f2. t + c2f2. .s + c2f2.C T 2 5 2 C I / I I S S + rScifi^s rc2f2 = c2f2.S(T))+.t) both satisfy Eq. t + -(r2S2c1fi!ss + rScifi. t) = 0..io)(o.t + \v2S2h. (10) At time t = T the payoff of the European put option will be max{0.ss rh = h. thus F(S.ss +rSc2f2.t) and f2(S.238 An Undergraduate Introduction to Financial Mathematics If the call option finishes out of the money. t) = K.T) = {K .t + \o2S2f2.25) = $0 . the investor exercises the put option and finishes with a portfolio worth again le(o. If the stock is worthless (i.t + 7jV2S2c2f2.25) + 3 0 _ 30e(o. (7.t + -^cr2S2c2f2tss r(cifi + c 2 / 2 ) = (ci/i.ss Therefore if c\ and c2 are real numbers rcifi = c i / M + .io)(o.s + rSc2f2.5) then + rSfhS + rSf2<s. K . 5 = 0) then the put option will be exercised and F(Q.S(T)} = (K S(T))+. 2 65862 > 0.s) + = (ci/i + c2h)t + \<?2S2{Clh c2f2)Ss +rS(c1f1+c2f2)s rf = ft + \cr2S2fss+rSfs.t) + ~cr2S2(c1fl!ss +rS(cifitS + c2f2.

a M < e~aM then by the Squeeze Theorem (Theorem 2.u ):.™ * dx pM Jo lim M^COJQ / e -(° +.: ' dx M 0 ~l lim e -( 0 +™)* a + iw M-^oo _ 1 lim ( e -(°+™)M a + 2W M-^oo Since 0 < | e .7 (1) Solution of the Black-Scholes Equation OO : / / dx -oo e .7 of [Smith and Minton (2002)]).^ e .Solutions to Chapter Exercises 239 B.™ M | e . lim (e-(*+iu)M (2) _A fH a + iw fix) = -^ J°° /Me™* dw 2TT e-awelwx M dw -(a-ix)™ d w — lim — 27T M ^ o c -1 -(o — ix)w — lim 27r M-^oo a — ix lim (e-(a-ix)M 27r(a — ix) M^OO M _ A Since 0 < \e~ixM\e-^aM < e~aM then by the Squeeze Theorem (Theorem 2. lim f e -(«-«) M _ A = _i => f(x) = 27r(a — ix) .7 of [Smith and Minton (2002)]).

240 An Undergraduate Introduction to Financial Mathematics (3) f(x) = ^ l 1 2^ f(^ywxdw e-aw &lWx dw -a(w*-i*w) d w 2W_.t)) = K^(v(x.r)) ( dv dx \dx dt /dv \dx Ka2 dv dv dr dr dt dv dr 2~lfr . 2TT !TT r J-oo r°° e- e-^^+i^e-^^dw J_e-*2/4a f -z2/4a 2TT 1 <w-£¥ dw dz / e~az- 1 -x2/4a 2\/Tra (4) ^(F(S.

kv ?V + Vxx -Vx + ~^Vx (e(k+l)(x+V2^y)/2 _ e{k~l)(x+^2^y)/2\+ > Q .l)vx .vx) + 2 _2a: Ka 1 e ff2 rv = ° —vT 2 + 2a ^Vxx ~~ Vx> + l 2. rVx Vr = az a2= -kv + vxx -vx + kvx = vxx + {k. (7.Solutions to Chapter Exercises d2F OS2 d (dF dS V dS d gg{e-xVx) _„ —e Vr x by Eq.+ vXT • 0 ^xa: J j f e x ^ g -2a. -vx) -Vx) rF = Ft + ^a2S2Fss 2 + rSFs rSe~xvx rKv = Ka 1 e~2x 2 2 — i v + ^ S —j-^{vxx . = -rv + -a (vxx .vx) + rvx 2r 2r ~YVT 1 2.11) he dx dS _T ( dx Vr-r \ xxdS h Vrr dr dS -e~xvx e e^ ^ - • — + e~x I vxx • .

31) becomes <vX/V2T (fc-l)y/2T1 ^2 2 n / ) T ) / 2( feci. . _ (/c ± l)x Thus the integral in Eq.l^^^ + (fc-l)x/2 dy {k + l)V2^^2 / ) V 02 e ^ 4 L J : + ( f e + 1 W2 d y (fc + 1) -{yC/V2T e(fc+l) 2 r/4+(fc+l)z/2 /2TT ^oo x/V2r -(y-^¥^f/2dy -(y(k-l)V2¥^2 = (fc-l)2r/4+(fc-l)z/2 V27T J-X/V2T r/2 dy -x/V^r (9) If we let w = -y + \{k + l)-v/2r. then dw = -dy.+V2?v)/2 _ p(k-l)(x+V2Ty)/2\+ -y2/2 'dy 2 (e{k+l)(x+-j2^y)/ 2 _ .V^(fc ± l)j/ + ^ ^— J .. (7.242 An Undergraduate Introduction to Financial Mathematics if and only if (fc+l)0+V^n/)/2 ^ ex+V2ry > e^ ^ ^ (fc-l)(:H-v/2?2/)/2 x + y/2ry > 0 y> I IT (8) By the result of exercise 7 we know that 1 /2TT ( > + D)(a.i ^ (fc±l)v / 2 : f\ (k±l)2 (k±l)x . When y = .y/2r(k ± l)y .(k ± l)x 2 /s-„ .l X x + V ^ M V27T J-X/V2T 0[(k+l)(x+V2^y)-y 2 p -y /2•dy 1 >2ll J-X/V2r ]/2 _ p[(k-l)(x + ^y)-y2}/2 dy Completing the square in the exponents present in the integrand yields y2 . ( f c . ^ (fc±l)2r\ (fc±l)2r y2 .

then w = x/y/2~r + |(fc — 1 1)V^T: then dw = -dy.Solutions to Chapter Exercises 243 then w - X/V2T+ \(k + l ) \ / 2 r and V27T J-x/y/2^ 27T / J — oo Jx/V^+^ik+l)^ -w2'2dw e-w'2dw {wA{k+l)^)' (10) If we let w = -y+\(k-l)y/2r.i ) ^ ^ / J —oo e-2/2dw e~w I2 dw (= j (11) !/. ^ '2 7 ^ ^= 2b ln(5/if) 1 /2r ln(S7in = + 1 /2r i /„ « / m ^ + lh/2 T(r t) ' - \ ^ aVT^t HS/K) ay/T-t ln(S/K) [ \a2 2t (^ + l)g2(r_t) ay/T-t a2/2)(T-t) + (r + ay/T-t . When y = and -xjyf^r /-oo 2 :L / e -(y-|(fc-i)v^) /2 d 2 / 27T J-x/^/2^ 1 /• — o o i= / 27r A / v ^ + ^ f c . ^ /TT(H1) + i(t-i)vS).

2.2 of [Smith and Minton (2002)]).6 2. (13) Using the parameter values specified w « 0.8 2. The graph below shows that the solution is near 0. . (14) One way to find the volatility is to approximate it using Newton's Method (Sec.50.244 An Undergraduate Introduction to Financial Mathematics 1 J IT & r ln{S/K) + (r + ay/T-t ln(S/K) + {r + IIT (fc + l ) V 2 r . We wish to find a root of the equation Ce(a) = 2. 3.40141.2 Newton's Method approximates the solution as a « 0.536471 and thus Ce « $5.V 2 r a2/2)(T-t) a2/2)(T-t) 2-y(T-t) WT-t (12) Using the parameter values specified w « 0.4 2.16662 and thus Pe w $6.827888.80 which we will use as the initial approximation to the solution for Newton's Method.05739.

Thus rf = ft + \cr2S2fss + rSfs rS = 0+ ]-a2S2 -0 + rS rS = rS. and ft = 0. fss = 0.Solutions to Chapter Exercises 245 (15) If f(S. (8. r = 0. t = 0. T = 1/4.t) = S. then fs = 1.1)) and Eq. (8.3).139819 6 = -33. Therefore the price of the security itself solve the Black-Scholes partial differential Eq. B.03. and a = 0.5). According to the Put/Call parity formula.8 Derivatives of Black-Scholes Option Prices (1) Using Eq.25 we have w = -0. (6.3) with S = 300. dP 3C ^ _rrr_0 Substituting the expression already found for ^ in the right-hand .281 (2) To find O for a European put option we will make use of the Put/Call parity formula (Eq. K = 310. (7.

4>{w CTVT -1)) ~ I2 c r ft' e -''(T-t)-(w-<Tv T^t) 2 /2 2^2n(T-t) r.w) fl^S/K) .t) + ( i .r = 0. t = 0. K = 265.t) \ T-t a2/2 2 + Kre-r(T-t)(j){aVT~^t Ke-r(T-t)-(w-*VT=tf/2 . T = 1/3.20 we have w = 0.w) \ T-t T + + Kre-r{T-t)4>{aVT~^t 2a^2ir(T .t) ° ' (3) Using Eq.r ( T . and a = 0.* ) ( r4>{w aVf~^t) I2 (HS/K) \ T-t ae-(w-aVT^iy/2 _ 2 2^2ir(T-t) (MSIK) \ T-t / J _ ~ r Se-^l2 - ge-r(r-t)-(M-ffy^)V2 2 a 2<ry/2Tv(T .35) HS/K) w + (r + a2/2){T-t) aVf~^t .3073 (4) According to Eq.5) with 5 = 275.0-72 .246 An Undergraduate Introduction to Financial Mathematics side of this equation yields dP _ Se~w I2 . (7.t) \ (HS/K) T-t r . (8.02.436257 6 = -15./ 2O^2TT{T w2 2 (HS/K) . r Se.Ke-rV-V-^-*^^ dt 2ay/2ir(T -t) . ' \ ) / 2^2TT(T -1) —7^r—k ~ \<j(T-t) ° -a ' 2 +a) Se~w2l2 2ay/2Tr{T .i f e .

(8. (8. K = 140. K = 165. r = 0.299167 (6) Using Eq.11) with S = 180. and (j = 0.30 we have w. T = 2/3.526797 A = 0. T = 1/3.8) with S = 150.321416 r = 0.15 we have w = -0. t = 0. = 0.35) w_ HS/K) + {r + ^/2){T-t) .055.0375.564706 A = -0.Solutions to Chapter Exercises 247 which implies that dw ~dS = _ ~ d \n{S/K) + {r + a2/2)(T-t) 'dS aVT^t _d_ ln(S/K) dS 1 d [HS/K)} aVT^tdS 1 d [In S .0121519 (8) According to Eq.713863 (7) Using Eq.22 we have w = -0. and a = 0.025. r = 0. r = 0.9) with S = 125. t = 0. T = 5/12.I n K] ay/T=tdS 1 1 ay/T^t S 1 aSVT^t (5) Using Eq. (8. and a = 0. (7. K = 175. t = 0.

r = 0.14) with S = 300. T = 1/3. K = 272." * ) . and a = 0.9247 (10) Using Eq.4071 (11) Using Eq.(HS/K) + (r + a2/2)(T-t)) a2VT^i 2 a {T-t) HS/K) + {r + a2/2)(T-t) a2y/T^t a2y/T=t 1 f\n(S/K) + (r + a2/2){T-t) y/T^t-.25 we have w = 0.129235 V = 83.0475.r = 0.0255.35 we have w = -0. r = 0.t))y/T -1 a2(T-t) a2{T.19) with S = 270.012164 p = -23.0758 . t = 0.15 we have w = 0. and a = 0. T = 1/2. (8. t = 0. T = 1/6. and a = 0..0375.t))y/T^t a2{T-t) 2 a2(T. K = 325. a \ ay/T^t (9) Using Eq.t)ay/T~^t .248 An Undergraduate Introduction to Financial Mathematics which implies that dw da 2 d ln(S/K) + (r + a /2)(T-t) da ay/T^t a{T .tf' -(ln(S/K) + (r + a2/2)(T. (8. (8.(ln{S/K) + (r + a2/2)(T .171209 p = 38. K = 305. t = 0.19) with S = 305.

Start by considering the following limit.Solutions to Chapter Exercises 249 B.V lim (WW t^T.\ay/T-t = +00 + 0 = + OO + a- ir^yi)^ We know the limit is +00 since \n(S/K) > 0 for an in the money call option.\ay/T^t = lim (Msm t^T. < 0 for an out of the money . lim t_T- „= lira t->T~ (MW+iL+f/HKIzi) OyjT -t (r + ay2)(T-t) (Ty/T-t V = l i m (HS/K) t^T. Start by considering the following limit. (2) If the call option is out of the money then S < K.\ay/T -t + (r+^/2)(T-t) ay/T -t 1 (r + v2/2)(T-t)\ (Jy/T-t J + = lim t-r= —00 (H^m+(r \ay/T-t aV2)vT^t = -OO + 0 We know the limit is —00 since \n(S/K) call option. n m u .9 Hedging (1) If the call option is in the money then S > K. Now since A c = 4>{w) an<A <fr(w) is a continuous function then lim A = lim <b{w) t->Tt->T- lim w t~>T~ = lim 4>(w) w—^oo = 1. = l i m t-T= (HS/K) t^T.

06 51.614837 0.954430 0.31 A 0.408940 0.782988 0.98 54.698867 0.55 47.824866 0.00 44.526428 0.58 46.79 48.23 47. (3) These calculations are similar to those used to create Table 9.583262 0.581970 0. 10 .995748 10 .986327 0.36 52. 80 72 104 115 122 116 159 168 129 142 156 199 206 208 210 0 .250 An Undergraduate Introduction to Financial Mathematics Now since A c = 4>(w) and </>(«.28 49.07 47.366642 0.761174 0.62 47. Shares Held 2045 1833 2632 2910 3074 2916 3915 4124 3177 3494 3806 4772 4932 4979 5000 5000 Interest Cost Cumulative Cost 92025 82654 119919 133152 141077 133729 183396 194019 148930 164379 179750 229520 238048 240697 242044 242254 ek 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 S 45.) is a continuous function then t->T- lim A = lim <j>(w) t->T- = <p I lim w V-T) = lim <j>(w) = 0.60 50.22 54.33 48.81 51.1.635479 0.

235189 0.52 41.77 39.254169 0.447836 0.408940 0. At the time the Put option is issued its price is $2. the put option finishes out of the money.114050 0.83 44. Delta for a European Put option is always negative.140594 0.64 44.90 43.42 42.001621 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 703 365 366 380 488 1271 1176 00 . In other words. thus the financial institution will take a short position in the stock in creating the hedge.94 42.097654 0.58 45.0 570 8 0 0 80 72 88 74 48 30 18 18 19 23 53 49 26 6 6 0 (5) In this case the put option will not be exercised since the final price of the stock exceeds the strike price.18 41.73256.073130 0.72 44.23 41. 0.93 44.073032 0.Solutions to Chapter Exercises 251 (4) These calculations are similar to those used to create Table 9.19 41.00 44. Shares Held 2045 1833 2239 1873 1191 Interest Cost Cumulative Cost 92025 82654 101255 84909 55370 34810 20921 20981 21586 26219 61343 57128 30399 6950 6640 6645 Week S 45.56 40.366642 0.62 A 0.045 yr _ 1 .238290 0. The revenue generated by the sale of the Put option and the short position in the stock will be invested in a bond at the risk-free interest rate r = 0.374621 0.2.075921 0. .

364521 -0.416738 -0.301133 -0.60 50. .79 48.252 An Undergraduate Introduction to Financial Mathematics ek 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 S 45.20.004252 Shares Shorted 2955 3167 2368 2090 1926 2084 1085 Interest Earned 876 1823 1506 1194 0 0 228 68 21 0 0 0 115 123 91 80 73 80 37 28 67 54 41 -2 -9 -11 -12 Cumulative Earnings 132975 142541 105471 92432 84703 92246 42776 32348 77633 62380 47205 -2368 -10699 -13152 -14302 -14314 Assuming the buyer of the option pays for the option at the strike time the net cost to the financial institution is (5000)(2.58 46.045570 -0.591060 -0.06 51.013673 -0.633358 -0.00 44.175134 -0. (6) In this exercise the Put option will be exercised since the final stock price is smaller than the strike price.23 47.81 51.238826 -0.07 47.73256) .385163 -0.217012 -0.36 52.31 A -0.98 54.418030 -0.55 47.28 49.62 47.22 54.473572 -0.33 48.14314 = -651.

§) = 2. | ) — WQC{S.2972 and C(85.859406 -0.62 A -0.19 41. Note that C(85.73256) = -43. This final transaction changes the holdings of the financial institution by 221294 . \) = 1.822155^6- .56 40.64 44.90 43.Solutions to Chapter Exercises 253 Week S 45.055.94 42.0. then the value of the portfolio is V = w±C(S.18 41.77 39. and a = 0.23 41.59313.998379 -1.764811 -0.0474901^4 .0 -1.885950 -0.52 41.926968 -0.591060 -0.924079 -0. r = 0. (7) Let S = 85.0 Shares Shorted 2955 3167 2761 3127 3809 4297 4635 4634 4620 4512 3729 3824 4430 4992 5000 5000 Interest Earned 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 0 115 123 107 122 148 166 178 178 177 174 143 147 171 191 191 Cumulative Earnings 132975 142541 124135 140676 170410 191166 205250 205386 204977 200540 165612 170024 196950 220595 221103 221294 At the strike time the financial institution must honor the Put option and cancel its short position in the stock.72 44.42 42.93 44. Thus the financial institution purchases 5000 shares of the stock at the strike price of $47 per share.0390443w6.552164 -0.83 44.625379 -0.20.926870 -0. | ) . Thus the portfolio is Gamma-neutral whenever W4 = 0.(5000)(47) + (5000)(2.58 45. We will let the number of four-month options be W4 and the number of six-month options be WQ.00 44.17.902346 -0.633358 -0. \iV represents the value of the portfolio consisting of a short position in the four-month options and a long position in the six-month options.745831 -0. K = 88.761710 -0. Gamma for the portfolio is r = T4W4 — YQWQ = 0.

If V represents the value of the portfolio consisting of a short position in the three-month options and a long position in the five-month options.496454W5 = x0.and Delta-neutral whenever x = 0. | ) — w$C(S.117028W5. \). Thus the portfolio is both Gamma. Delta for the portfolio is A = x + A4W4 — AeWQ = x + 0.6 .4 . JTJ). ^ ) . \) = 1. ^ ) = 3. We will let the number of threemonth options be u>s and the number of five-month options be W5.74863 and C(95.489693) (0.0.0365043w.0.500131w6 = x .T 5 w. Gamma for the portfolio is T = T3W3.045.0. | ) — WQC(S.774825)w5 .0. and a = 0.5 = 0.0. r = 0.127514u>6.0. . the value of the portfolio is now V = Sx + WsC(S.and Delta-neutral whenever x = 0.254 An Undergraduate Introduction to Financial Mathematics (8) Using the results of exercise 7 and including a position of x shares of the security.127514w6.489693^3 . Thus the portfolio is both Gamma. Thus the portfolio is Gamma-neutral whenever W3 = 0. the value of the portfolio is now V = Sx + WiC{S.45322w.496454w5 = x + (0.45322)(0. then the value of the portfolio is V = u>3C(S.3 . Delta for the portfolio is A = x + A3W3 A5w5 = x + 0. Note that C(95.08418.117028^5- .23.0282844w5.822155)u. (10) Using the results of exercise 9 and including a position of x shares of the security.7748251^5.500131™6 = x + (0. (9) Let S = 95. | ) — w$C(S.

2)) 2 2 (by Eq. if X and Y are independent random variables Var (X + Y) = Var (X) + Var (Y).X) (by Eq. X + Y) = E [(X + F ) ( X + Y)] .E [X] E [Y] = 0. In other words X = 0 which implied XY = 0 and thus 0 = (E [XY]f < E [X 2 ] E [y 2 ] = 0. then by Theorem 10.10 Optimizing Portfolios (1) Proof of statement 1: Cov (X. Y).(E [X])2 .(E [X])2 + E [y 2 ] .2E [X] E [Y] .2)) (by Eq.(E [Y]f = E [X 2 ] . X) = E [X • X] . (3) Suppose that E [X 2 ] = 0.E [Y] E [X] = Cov(y. then P (X = Xi) = 0.(E [X] + E [Y]) (E [X] + E [Y]) = E [X 2 ] + 2E [XY] + E [y 2 ] .E [X + Y] E [X + Y] = E [X 2 + 2XY + Y2] .2)) (by Theorem (2. Recall that if X and Y are independent random variables E [XY] = E [X] E [Y] and thus E [XY] . .6)) (2) Let X and Y be random variables.E [X] E [Y] = E [YX] . Therefore. Therefore we must have P (X = 0) = 1.(E [Y])2 + 2(E[Xy]-E[X]E[y]) = Var (X) + Var (Y) + 2Cov (X. (10.1 Var (X + Y) = Cov (X + Y. then by definition V[X2]=J2*i-V(X i=l = Xi)- If Xi ^ 0 for some i. (10.Solutions to Chapter Exercises 255 B.E [X] E [X] = E[X ]-(E[X]) = Var (X) Proof of statement 2: Cov (X. (10. Y) = E [XY] .

9923)(56. This quadratic function is minimized when n = 1.045)2 = 0.5641) (5) If n is the ratio of the number of shares of security D purchased compared to the number of call options C sold. (6) According to Theorem 10.0.069.6923 E[XY] =671.045)(0.9923 E [Y] = 56.80207 p{x. D) + Var (C) = (0.002025. then (E[Xy])2<oo = E [ X 2 ] E [ y 2 ] . (4) Let X be the golf handicap of the CEO and let Y be the stock rating of the CEO's corporation. ~8-8°207 = = -0. ^(6.(11. Suppose that E [X2] = oo.04595.4.8741 Var (Y) = 45.2nCov (C.Y) = .5641 Thus we have Cov(X. is concave when u"(x) < 0. then the value of the portfolio is V = C — nD which implies the variance in the value of the portfolio is Var (P) = n 2 Var (D) . u'{x) = x «"(*) = "^ < ° for 0 < x < oo.6923) = -8.256 An Undergraduate Introduction to Financial Mathematics A similar statement can be made when E [Y2] = 0.037) 2 n 2 .069 Var (X) = 6. .001369n2 . a twice differentiable function u(x).8741) (45.Y) = 671.497354. A similar statement can be made when E [Y2] = oo.0028638n+ 0.037) + (0.86)(0. Then we may calculate the following quantities: E[X] = 11. (a) u(x) = In a.2n(0.

. It is concave only for x > 0. u (x) u"(x) 2 lnx x 2(1 . .1 x u'{x) u (x) = 1 1 + x2 (1 + x 2 ) 2 This function is not concave on its whole domain. in other words only for x > e.l n x ) This function is not concave on its whole domain.. (7) According to Jensen's inequality (10. It is concave only for x such that lnx > 1. (d) u(x) = t a n . since f(x) = lnx is a concave function then ln(^XP(x)) > J>XP(X) Vx J x = ]Tln(xpW) x = ln (ll xPm )Since the exponential function is monotone then E [X] = e l n & * xp^) > e l n (n* * P < X ) ) = Q.Solutions to Chapter Exercises 257 (b) u(x) = ln(x 2 ) Since ln(x 2 ) = 2 lnx. (c) u{x) = (lnx) 2 .14). then by the previous result ln(x 2 ) is concave for 0 < x < oo.

258 An Undergraduate Introduction to Financial Mathematics Now if X > 0 then T = ^ > 0. Thus the following string of inequalities is true. an obvious one. V Ml " x 2-^x x Again equality holds if and only if X takes on only one value. Then we have X = H = Q = E [X]. Now suppose that all the random variable X takes on only one value (with probability 1). ^ I P ( I ) < m f { I } x Equality holds if and only if X takes on only one value. From this inequality follows Y ^ r 1 ^ X < m a x {l . By the previous result l[T^xUeE[T] x = E T P PO X Ylxxnx) x < -£ ^ l^x P(X) x x Q>n.x{X} = m a x x { I } and thus min{X} = H = Q = E [X] = max{X} X -A .3 } x X E[X] = > min{X}. Thus we have shown that TL < Q < E [X]. Before finishing the exercise we must establish another inequality. min{X} < H < G < E [X] < max{X} yC -A Now if m i n x { ^ } = 'H then X takes on only one value and we have mm.

— + 2 V 50 (~2) 2 50 = c. i.3\/53 w 3.tdt Jo = ln(cosh£)| 0 = ln(coshl) « 0.15967. f JO f(<p(t)) dt= I t&nb.Solutions to Chapter Exercises 259 (8) <j>{t) = t.433781 < 0.— 50 C2 50 r2 r2 50 52 25 25 C = 25 ± 3V53 The certainty equivalent will be the smaller of the two roots of the quadratic equation.462117 tanh tanh tanh / % ( * ) dt Jo = ln(coshl) ..ln(coshO) (9) The certainty equivalent is the solution C of the equation 2(/(10) + / ( " 2 ) ) = / ( C ) 102 -I 1 0 . C — 25 .e.— = c. .

i.l . C = 2 ~ v 2 1074 « —15. l l ( l This function is maximized when a)x). then u'{x) = be"hx and u"{x) = -b2e~bx.11(1 .e~bx where 6 > 0. (11) If we let X represent the random variable of the investor's return on the investment. \ 1.e.p ) u ( l . The function u(x) is monotone increasing since u'(x) > 0 for all x.386.260 An Undergraduate Introduction to Financial Mathematics (10) The certainty equivalent is the solution C of the equation 5/(15) + | / ( . Since 6 > 0 then -b2 < 0 and thus u"(x) < 0 for all x.011236(200p-lll) (l-p)(-l.) 1.11(1 . then the expected value of the utility function is E [u(X)\ = pu(2ax + 1. o = AE[ttW] p(2z . Therefore if 0 < x\ < Xi then u{xi) < u(x2)- .11(1 — a)x with probability 1 — p If the investor's utility function is u(x) — \nx.11(1 — a)x with probability p.a)x) + (1 .. then J 2ax + 1. 2 /-255\ 3 \ 2 j 2 /-255\ 3 V 2 y 535 _ 2 C2 . l l z ) 2ax + 1.11(1 -a)x a = 0.2 C2 2 C2 Q — C 2±V1074 " 2 The certainty equivalent will be the smaller of the two roots of the quadratic equation. Hence u(x) is concave.ax) (12) Let u{x) = 1 .lla.1 5 ) = / ( C ) 3\ 2 J 3\ 1 /-195\ 3\ 2 / 1 /-195\ 3 V 2 7 2 ) .

17.41 ~ 0.26) = 0. E [W] = 1000 + 0.33 OLE 1 0JS3 0. (by Eq.j/)(0.27 ~ 0.03802y + 1089.24 ~ 0.7 aA = aB l 0. E [W] .y) = 1130-0.212697 = 0.319045 = 0.24 ~ 0.Var (W) /200 is maximized.05y Var(W) = (0.41 ~ 0.Var (W) /200 = -0.41 """ 0.24 ~ 0.27 "• 0. 1_ 0.17)) = E cy^jjJi(Ri = E[cR(w)] = cE [-R(w)] = cr(w).124505 = 0.16 ' 0.16 ' 0.604y + 8100 The investor's utility function u(x) = 1 — e ~ x / 1 0 0 .3) .27 ^ 0.27 ~ 0.436.13(1000 .24 ~ 0.5 This quadratic expression is maximized when y « 365.0% + 0.33 1 0.154689 0.Solutions to Chapter Exercises 261 (13) Let W represent the wealth returned to the investor and let y represent the amount of money invested in the first investment (0 < y < 1000). is maximized when E [W] .27 cue 1 0. (10.16 ~r 0.16 ' 0.17)) (by Theorem 2.03)(0.33 1 0A1 0.16 a D = i 1 _1 | 1 . (10.189064 = 0.33 1 0.03) V + (0.010404y2 .00005202y 2 + 0.16 ' 0.41 ~ 0.09)2(1000 . (14) According to Theorem 10.33 (15) If c is a real number then r(cw) = E [R(cw) :E ^2cu>i(Ri i=l -r) -r) (by Eq.09)(-0.27 ' 0.24 "•" 0.24 = 0.41 ~ 0.y)2 + 2y(1000 .

2) (16) In this example. . n. V(cr 2 (w)) = A V ( r ( w ) ) r(w) = 1 The set of equations above are equivalent to the following set of n + 1 equations.r) ] (by Eq.. . 2 . (10.17)) (by exercise 18 of Chap. ..17)) Var \ t=i Ic^MRi-r) / = Var (cR(w)) = c2Var (R(w)) = cV(w) (by Eq. 2wi(if = \{ri — r) n for i = 1. Ywi(ri 1=1 ~r) = l If we solve the ith equation for Wi for i = 1.n and substitute into the last equation we have (ri .. We must solve the following system of two simultaneous equations. . the variance in the rate of return is to be minimized such to the constraint that the expected value of the rate of return must be one. . (10. We will again use Lagrange multipliers to find the minimum variance.2.262 An Undergraduate Introduction to Financial Mathematics Likewise 2 CT (cw) = Var(i?(cw)) = Var [Y^cwiiRi vi=l .r) 1=1 This implies that {ri—r)2 2a'2 En i=l .

rf) "M n . r = 0. T = 1/4.0.15 into Eq.Solutions to Chapter Exercises 263 Therefore 2 v-^" (rj-r) 2 9^. and K = 86 we can determine the value of the European call option as C = 1.69925 (18) We must show that the set of equations below is satisfied by the expressions in Eqs.055.0475.137376 (20) Using the values 5(0) = 83.9031 a D = 0.13. n . \i = 0. and Cav(i. (10.23423s + 3.2piimo-io-M + c ^ ) ] ) a2 = x2a2 + (1 .0475 + (3.09917)(0.rM = A (2 [{pt^MCfi .0475) n = 0.3 aA = 4.0765.26152.0 aE = 9.2 V^« ( n . a = 0.r j " zo 2oT i 2 ^ = 1 2o-2 ^ r a (ji_ v^n Zw=l (17) According to Lemma 10.25. (10.31) the expected return of the portfolio is a linear function of x and y given by E [R] = 2.18342y.28) and (10.32). According to Eq. t = 0.29).a-M)(TM + x(cr2 .22.0475 = 7 ^ 5 ( 0 . rM = 0.M) 0.30112 ac = 7.44549 a B = 1. The variance in the expected rate of return is a quadratic function of x and y described in detail by Eq.029) = 0. -J ( r M .x)2a2M + 2x(l x)piMOiOM = (19) Substituting rf = 0. (10. aM = 0. Suppose a portfolio consists of a position x in the security and a position y in the option.0. M) . (10.0 7 6 5 . This can be accomplished by direct substitution into the equations. In this example the .26) we obtain n-rf = Cov (i.

701834a.64607. Substituting this expression into the variance of R produces the quadratic function of x given below. We would like to find the minimum variance portfolio having E [R] = 10.264 An Undergraduate Introduction to Financial Mathematics quadratic function takes the form Var (R) = 114. .727 The minimum value occurs when x = —2.6908y2.2 + 129.. Var (R) = 46.2 + 200. If we set the expected rate of return to $10 and solve for y we obtain y = 3. + 460.264xy + 46.14408 and thus y = 4.6819a.14128 -0.405a.179a.

(2001). D. Johns Hopkins University Press. C. J. and Faires. Brief Calculus and Its Applications. Wiley. D. R. and Robbins. C. pp. D. 180-182. Boyce. (1975). Available at http://www3. CRC Press. W. W. F. Basic Partial Differential Equations. R. G. USA. 265 . I. and Verhey. and DiPrima. Prentice-Hall. NY. Upper Saddle River.. Baltimore. D. (1996). W. New Jersey. J. UK. J. Brown. Activity Analysis of Production and Allocation. International Press.txt. Oxford. W. 3rd edition. Dean. Prentice Hall. New York. (1976). USA. (1996). List of Playboy's playmates of the month with data sheet stats. D. and Tucker. Burden. Goldstein. and Schneider. USA. R. 7th edition. NJ. DeGroot.. Gardner.ca/jimdean/pmstats. and Ewell. Inc. V. Courant.C. and Van Loan. G. R. A. D. USA. (2005). (2001). 2nd edition. What is Mathematics?. (1999). McGraw-Hill Book Company. Cambridge. H. Koopmans. Massachusetts. USA. Lay. New York. Reading. Advanced Engineering Mathematics. M. pp. Massachusetts. Behavioral Science: Quantitative Methods. (1969). FL. Oxford University Press. (1951). CA. Churchill. 2nd edition.. 317-329. Upper Saddle River. USA. Boca Raton. Gale. C. Addison-Wesley Publishing Company. J. H. Linear programming and the theory of games. Maryland. Massachusetts. USA. Stochastic Calculus: A Practical Introduction.sympatico. John Wiley and Sons. M. R. (1998). Numerical Analysis. 8th edition. L. Elementary Differential Equations and Boundary Value Problems. New York. Kuhn. E. Advanced Engineering Mathematics. Thomson Brooks/Cole. (2002). H. A. H. Corvin. Burlington. Belmont. Scientific American. Jeffrey. and Csordas. Durrett. Matrix Computations. M. Golub. Probability and Statistics. editor. 8th edition. F.Bibliography Bleecker. R. USA. L. Harcourt Academic Press.. USA.. USA. D. Greenberg. "Mathematical Games" column. J. Complex Variables and Applications. in T. USA.. (1989). (October 1959). Inc.

Elementary Stochastic Calculus With Finance in View. 2nd edition. Advanced Calculus. E. S. R. NY. 12. Cambridge. Vol. American Statistician. and Mann. J. "Ask Marilyn" column. (2002). E. USA. S. Freeman and Company. Inc. 7th edition. 1. Taylor. (2001). CA. Neftci. UK. Wellesley. The Mathematics of Financial Derivatives: A Student Introduction. UK. W. Springer-Verlag. 67. Cambridge. Oxford University Press. G. (1987). A. (February 1990). Cambridge University Press. G. New York. p. MA. M. and Dewynne. A problem in probability. CA. (2001). M.266 An Undergraduate Introduction to Financial Mathematics Kijima. Upper Saddle River. T. and Hoffman. Boca Raton. (2000). USA.. S. 3rd edition. Stochastic Processes with Applications to Finance. (2003). USA. Inc. Pacific Grove. USA. P. USA. J. USA.. USA. McGraw-Hill. FL. Marsden. and Minton. Introduction to Applied Mathematics. Chapman & Hall/CRC. Wilmott. W. J. Volume 45 in Applications of Mathematics. S. M. M. Selvin. New Jersey. Strang. A Guide to First Passage Processes. Steele. Parade Magazine. USA. Boston. (1983). M. New York. Basic Complex Analysis. 4th edition. Academic Press. B. Cambridge University Press. Calculus. (2006). Mikosch. R. MA. T. Prentice Hall. Investment Science. . 29. H. vos Savant. New York. NJ. Cambridge University Press. S. p. D. N. (1999). Ross. 6 of Advanced Series on Statistical Science & Applied Probability.. (1995). USA. R. S. An Introduction to Mathematical Finance: Options and Other Topics. J. Ross. (1998). UK. Stochastic Calculus and Financial Applications. USA. (1986). New York. M. Cambridge. Brooks/Cole Publishing Company. John Wiley & Sons. USA. J. Howison. NY. San Diego. World Scientific Publishing. NY. (1999). (1998). (1975). A First Course in Probability. Smith. Luenberger. Stewart. Wellesley-Cambridge Press. River Edge. An Introduction to the Mathematics of Financial Derivatives. Redner. 2nd edition. Calculus.

5 convex set. 115 binomial random variable. 184 267 CAPM. 179 constraints. 63. 145 Delta neutrality. 45 Central Limit Theorem. 40 American option. 93 continuously compounded interest. 65 continuous distribution. 95 compound amount. 167 conditional exit time. 132 multivariable. 2 compound interest. 145. 103 annuity. 72. 3 concavity. 18 additivity. 65 equality. 48 Delta. 80. 90 conditional probability. 110 Arbitrage Theorem. 23. 65 inequality. 79 beta. 36 continuous random variable. 42 Bernoulli random variable. 72 binary model. 133. 80 European call option. 155 cumulative distribution function. 110 boundary condition.Index absorbing boundary condition. 150 . 19 constrained optimization. 155. 112 absorbing. 99 closing price. 89. 42 Black-Scholes equation. 65 correlation. 182. 184 betting strategy. 103 Capital Asset Pricing Model. 113 Addition rule. 65 covariance. 87. 107 binomial model. 36 continuous probability. 49. 35 continuous random walk. 112 backwards substitution. 135. 183 capital market line. 63. see Capital Asset Pricing Model centered difference formula. 187 call option. 104. 36 arithmetic mean. 192 augmented matrix. 87 average. 98. 176 certainty equivalent. 23. 113 Brownian motion. 159 cost function. 104 area as probability. 89 bell curve. 91 continuous stochastic process. see expected value backwards parabolic equation. 172 chain rule. 3 compounding period. 8 arbitrage.

112 financial derivative. 15. 65 Euler Identity. 184 exercise time. 116 function concave. 112 in the money. 96 differentiation chain rule. 146 hedging. 110 finite differences. 120 hedge and forget. 111 European call option. 17 excess rate of return. 119 definition. 110 deterministic process. 103 exponential decay. 135. 98 discrete outcome. 47 experiment. 131 harmonic mean. 10 rate of. 91 . 21 independent random variables.268 An Undergraduate Introduction to Financial Mathematics Extreme Value Theorem. 103 call flnal condition. 94. 163. 17 independent. 38 of normal random variable. 65 inflation. 166 Fundamental Theorem of Calculus. 180 expected utility. 10 initial condition. 90 expected rate of return. 111. 116 of a convolution. 167 utility. 87 generalized Wiener process. 40 derivative. 115 alternative definition. 179 Greeks. 71 effective interest rate. 45 Fourier Convolution. 84 differential equation stochastic. 136. 116 European call option boundary condition. 3 dependent random variables. 143. 16 distribution continuous. 36 distribution function. 192 geometric series. 123 exponential growth. 171 expected value. 113 European option. 15 certain. 187 dual. 132 future value. 7 gradient operator. 97 geometric mean. 111 interest. 148. 21 mutually exclusive. 65 final condition. 16 events compound. 5 Gamma. 3 efficient markets. 118 of a derivative. 117 frequency. 112 event. 63 equality constraints. 67 Duality Theorem. 131 financial. 16 discrete. 103 exit time. 65. 51. 151 Gaussian elimination. 40 inequality constraints. 1 compound. 41 difference equation. 24. 151 Gamma neutrality. 15 expiry date. 112. 153 independent events. 192 heat equation. 15 impossible. 117 Fourier Transform. 5. 92 deviation. 178 feasible vector. 36 drift. 164 hockey stick. 115 existence.

103 option pricing formula European call. 111 second order. 16 classical. see principal amount principal amount. 104 positive part. 65 optimal portfolio. 87 mean. 173. 26. 140 loans. 125 European put option. 19 continuous. 164 option. 16 conditional. 170 joint probability function. 42. 100. 104. 125 out of the money. 171 Discrete Version. 48 normal random variable. 2 interest rate. 39 Lagrange Multipliers. 163. 18 Multiplication rule. 151 optimal. see expected value Mean Value Theorem. 179. 125 primal. 103 European. 104 short. 111 parabolic type. 163. 92 normal probability distribution. 103 American. 111 Black-Scholes. 5 effective rate. 125 European put. 97 Ito's Lemma. 165. 180 position long. 162 Delta-neutral. 171 Continuous Version. 164 rebalancing. 145 Gamma neutral. 65 linear systems tridiagonal. 16 marginal. 46 portfolio. 15 discrete. 55 present value. 67 principal. 146 portfolio selection problem. 5. 84 linear programming. 119 investor risk-averse. 104. 87 linearity. 187 Jensen's Inequality. 20 noise. 3 simple. 188 long position. 185 Laplacian equation. 64 operations research. 110. 27 probability distribution. 169 minimum variance.Index continuously compounded. 7 lognormal random variable. 1. 177 Monty Hall problem. 48 . see expected value. 2 probability. 175 Portfolio Separation Theorem. 170. see partial differential equation Poisson equation. 139 linear. 36 empirical. 35. 21 normal. 168 Ito process. 103 put. 1 Inverse Fourier Transform. 110 pricing formula European call option. 27 matrix upper triangular. 51. 103 call. 187 marginal distribution. 111 PDE. 15. 140. 112. 168 risk-neutral. 86 polar coordinates. 42 odds. 153 outcome. 40 marginal probability. 16 269 partial differential equation.

159. 187 Weak Duality Theorem. 167 variance. 87 uncorrelated random variable. 92. 35. 178 uniform. 15 lognormal. 41 alternative formula. 111. 1. 87 utility. 103 Put-Call Parity. 8 return rate of. 42 uncorrelated. 92. 96 stochastic process. 187 simple interest. 65 spatial homogeneity. 94. 150 Taylor's Theorem. 166 decreasing. 166 random variable. 49 Schwarz Inequality. 103 strike time. 163 of normal random variable. 41 minimum. 185 risk-averse investor. 83 strike price. 37 random walk. 162 excess. 167 utility function. 29. 155. 15. 11. 15 Bernoulli. 104. 180 rebalancing. 187 probability distribution function. 87 tridiagonal matrix. 47 vector inequalities. see rate of return. 51. 51. 77 continuous. 180 variance. 77 stochastic differential equation. 15. 164 symmetric random walk. 168 risk-neutral investor. 41 standard normal random variable.270 An Undergraduate Introduction to Financial Mathematics spread. 82 . 146 retirement. 131 Treasury Bonds. 48 Stirling's formula. 103 surrogate. 123 short position. 93 stochastic processes. 37 upper triangular matrix. 168 samples means. 184 expected value. 163. 96 continuous. 137. 136 volatility. 2 slack variables. 42 continuous. 168 sensitivity. 110 stopping time. 11 Rho. 68 Wiener process. 44 stochastic calculus. 91 rate of inflation. 131 separation of variables. 66 Vega. 163 normal. 79 binomial. 134 random experiment. 160 secant lines. 78 Taylor remainder. 96. 98 Theta. 97. 10 rate of return. 159 uniform random variable. 105 tridiagonal linear systems. 106 Put-Call Parity Formula. 99 Taylor series. 35 discrete. 138 risk. 166. see standard deviation standard deviation. 36 put option. 32.

An Undergraduate Introduction to F i n a n c i a l --w Mathematics students who have completed a three or four semester sequence of calculus courses. It introduces the theory of interest. and portfolio optimization. hedging. stochastic processes.The student progresses from knowing only elementary calculus to understanding the derivation and solution of the Black-Scholes partial differential equation and its solutions. This is one of the few books on the subject of financial mathematics which is accessible to undergraduates having only a thorough grounding in elementary calculus. a r b i t r a g e .com . T his textbook provides an introduction to financial mathematics and financial engineering f o r undergraduate 6009 he ISBN 981 256-637-6 orld Scientific YEARS OF PUBLISHING 3 1 . It explains the subject matter without "hand waving" arguments and includes numerous examples. Every chapter concludes with a set of exercises which test the chapter's concepts and fill in details of derivations.2 o o 6 www.worldscientific. random variables and probability. option p r i c i n g .

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue reading from where you left off, or restart the preview.

scribd