Graduate Texts in Mathematics 1 TaxsvnZanio. ntrauction 49 Axiomatic Set Theory. 2nd ed (Oxroey, Messure and Category, 2nd ScnatrreR. Topological Vector Spaces. HurowStavbact. A Course in Homological Algebra. Mactase. Categorie forthe Working Mathematician. Hocits Pure, Projectve Planes ‘SERRE. A Course in Anhmetic TaxevrvZanise, Axiomatic Set Theory Hewes Introduction o Lie Algebras and Representation Theory. Cones. A Course in Simple Homotopy Theory. ‘Cowway. Functions of One Complex Variable. 2nd ed Beats. Advanced Mathematical Analysis “AnoERSONFULLER. Rings and Culepores of Modules, Gocunrsky/Gunsss Slble Mappings and Their Singularities Benaeaian. Lectures in Functional Analysis and Operator Theory Westen, The Sinicture of Fields. Rosenn.arr. Random Processes. 2nd ed. Hatstos. Measure Theory Hass. A Hilber Space Problem Book. 2nd ed, revise. Husewot.sn. Fibre Bundles. 2a Hunarinevs. Linar Algebra Groups. Banwes/Mack. An Algebraic Inoduction to Mathematical Logic CGneva. Linear Algebra. ath ed. Houses. Geometric Functional Analysis and its Application 25 HwereSmowaexa. Real and Abstract Analysis 26 Masts. Algebraic Theories 27 Keuzy. Genecal Topology 28 ZanisnuSasr. Commutative Algebra, Vo. 29 ZaniskuSnueL. Commutative Algebra. Vol. 30 Jacoeson, Lectures in Abstract Algebra I: Basic Conéeps 31 Iacossow, Lectures in Absract Algebra I: Linear Algebra. 32 Jaconson. Lectures in Abstract Algebea IL: Theory of Fields and Galois Theory. 38 Hascn. Differential Topology. 24 Srrzs. Principles of Random Walk. 2nd ed 235 Wenwer, Banach Algebras and Several Complex Variables. 2nd ed 36 Keutev/Nawioxa etal. Linear Topological Spaces 37 Movs. Mathematical Logie. 38 GRAuExr/PRTZSCHE. Several Complex Variables 39 Aaveson. An Invitation to C*-Algebns, 40 Kewesy/Saeuu/Knarr. Denumerable Markov Chains. 2nd ed ‘41 AvosTo.. Modular Functions end Dirichlet Series in Number Theory 42 Sean. Linear Representations of Finite Groups 43 Guumwwensox. Rings of Contintous Functions. ‘44 Kewoo. Elementary Algebraic Geometry. 45 Lotve Probability Theory 1. 4th ed 46 Lotve.Probabitty Theory I. th ed 47 Mass. Geometric Topology in Dimensions 2 and 3 coined i des Joannis Karatzas Steven E. Shreve Brownian Motion and Stochastic Calculus With 10 Illustrations Springer-Verlag New York Berlin Heidelberg London Paris TokyoAoannis Karateas Steven F. Department of Statistics Department of Mathematics Columbia University Carnegie Mellon University New York, NY 10027 Pinisburgh, PA 15213 USA USA Eduorial Board P.R. Halmos: E,W, Gehring Department of Mather Department of Mathematics Santa Clara University University of Mic Santa Clara, CA 95053 ‘Ann Arbor, MI 48109 USA USA [AMS Clawsieations: 6007, 6105 Library of Congress Cataloging in-Publication Data Karatas,loannis. Brownian motion and stochastic calculus. (Graduate texisin mathematics: 113) Includes bibliographies and index. 1. Brownian motion processes. 2. Stochastic analysis. Shrove, Steven EI Tile ML Series. QA247SKIT 1987 531.163 87-1281, {© 1988 by Springer-Verlag New York Ine. All rights reserved, This work may not be translated or copied in whole or in part without the writen permission of the publisher (Springer-Verlg, 175 Fifth Avenue, New York, New York 10010, USA), except for bref excepts in connection with reviews or scholarly analysis. Use in ‘onnestion with any form of information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed is forbidden, ‘The use of general descriptive names, trade names, tidemarks, ele. inthis publication, even if {he former are not especially denied, snot toe taken a8 sgn that such names, as understood by the Trade Marks and Merchandise Marks Act, may accordingly be used freely by anyone Lid, Hong Kong. Printed and bound by R. R. Donneley and Sons, Hat Printed in the United Stats of America. ironburg, Vir 987654321 (son oars Sing Jerlag New York Berlin Heidelber BN 3-540-96835-1 Springer . erlag Bevin Wopdabneaus or To our wives, Eleni and DotPreface Two of the most fundamental concepts in the theory of stochastic processes are the Markov property and the martingale property.” This book is written for readers who are acquainted with both of these ideas in the discrete-time setting, and who now wish to explore stochastic processes in their continuous time context. It has been our goal to write w systematic and thorough exposi- tion of this subject, leading in many instances to the frontiers of knowledge At the same time, we have endeavored to keep the mathematical prerequisites as low as possible, namely, knowledge of measure-theoretic probability and some familiarity with discrete-time processes. The vehicle we have chosen for this task is Brownian motion, which we present as the canonical example of both a Markov process and a martingale. We support this point of view by showing how, by means of stochastic integration and random time change, all continuous-path martingales and a multitude of continuous-path Markov Processes can be represented in terms of Brownian motion, This approach forces us to leave aside those processes which do not have continuous paths. Thus, the Poisson process is not a primary object of study, although it is developed in Chapter | to be used as tool when we later study passage times and local time of Brownian motion. The text is organized as follows: Chapter 1 presents the basic properties of ‘martingales, as they are used throughout the book. In particular, we generalize fom the diserete to the continuous-time context the martingale convergence theorem, the optional sampling theorem, andthe Dood Meser desomPoss tion. The latter gives conditions under which a submartingale can cationarity atthe ony thee + According to M, Lave “matingls, Maroy dependence dation at ten es fepzndence concept 0a ated whish ae sins ene a TS vaca tigation, yet with a great number of deep properties” (Amvii Preface as the sum ofa martingale and an inereasing process, and associates to every ‘martingale with continuous paths a “quadratic variation process.” This pro- cess is instrumental in the construction of stochastic integrals with respect to continuous martingales Chapter 2 contains three different constructions of Brownian motion, as well as discussions of the Markov and strong Markov properties for continuous-time processes. These properties are motivated by d-dimensional Brownian motion, but are developed in complete generality. This chapter also contains a careful discussion of the various filtrations commonly associated with Brownian motion. In Section 28 the strong Markov property is applied to a study of one-dimensional Brownian motion on a half-line, and on a bounded interval with absorption and reflection at the endpoints. Many densities involving first passage times, last exit times, absorbed Brownian ‘motion, and reflected Brownian motion are explicitly computed, Section 2.9 is devoted to a study of sample path properties of Brownian motion. Results found in most texts on this subject are included, and in addition to these, a ‘complete proof of the Lévy modulus of continuity is provided. ‘The theory of stochastic integration with respect to continuous martingales is developed in Chapter 3. We follow a middle path between the original constructions of stochastic integrals with respect to Brownian motion and the more recent theory of stochastic integration with respect to right-continuous martingales. By avoiding discontinuous martingales, we obviate the need to introduce the concept of predictability and the associated, highly technical, measure-theoretic machinery. On the other hand, it requires little extra effort to consider integrals with respect to continuous martingales rather than merely Brownian motion, The remainder of Chapter 3 is a testimony to the Power of this more general approach; in particular, it leads to strong theorems ‘concerning representations of continuous martingales in terms of Brownian motion (Section 3.4). In Section 3.3 we develop the chain rule for stochastic calculus, commonly known as 1t6's formula, The Gitsanov Theorem of Sec- tion 3.5 provides a method of changing probability measures so as to alter the drift of a stochastic process. It has become an indispensable method for constructing solutions of stochastic differential equations (Section 5.3) and is also very important in stochastic control (eg., Section 58) and filtering, Local time is introduced in Sections 3.6 and 3.7, and it is shown how this concept leads to a general Chapter 4 is a digression on the connections between Brownian motion, Taplace’s equation, and the heat equation. Sharp existence and uniqueness theorems for both these equations are provided by probabilistic methods: applications to the computation of boundary crossing probabilities are dis. cussed, and the formulas of Feynman and Kac are established, Chapter 5 returns to our main theme of stochastic integr ential equatio n and differ- ns. In this chapter, stochastic differential equations are driven Preface by Brownian motion and the notions of strony and weak solutions axe including comparison and approximation results, are offered tee ees Watanabe. Essentially equivalent to the search for a weak solution a tke In the context of this martingale problem, a full discussion of existence, uniqueness, and the strong Markov property for solutions of stochastic dik, ential equations is given in Section 5.4. For one-dimensional equations tg posible to provide a complete characterization af solutions which eit only upto an “explosion time,” and thisis set forth in Section 5 This section al presents the recent and quite striking results of Engelbert & Schmidt con, cerning existence and uniqueness of solutions to onedimensional equations This theory makes substantial use of the local time material of Sections A, 37 and the martingale representation results of Subsections 344. By analogy with Chapter 4, we discuss in Section 5.7 the connections between solutions to stochasti differential equations and elliptic and parabolic differential equations. Applications of many of the ideas in Chapters 3 and 3 are contained in Section 5.8, where we discuss questions of option pricing and optimal portfolio/consumption management. In particular, the Gitsanov theorem is used to remove the difference between average rates of return of different stocks, a martingale representation result provides the optimal portfolio process, and stochastic representations of solutions to partial die ential equations allow us to recast the optimal portfolio and consumption management problem in terms of two linear parabolic partial differential equations, for which explicit solutions are provided. Chapter 6 is for the most part derived from Paul Lévy’s profound study of Brownian excursions. Lévy’s intuitive work has now been formalized by such notions as filtrations, stopping times, and Poisson random measures, but the remarkable faet remains that he was able, 40 years ago and working without these tools, to penetrate into the fine structure of the Brownian path and to inspire all the subsequent research on these matters until today. Inthe sprit of Lévy’s work, we show in Section 62 that when one travels along the Brownian path with a clock run by the local time, the numberof excursions away from the origin that one encounters, whose duration exceeds a speci umber, hasa Poisson distribution, Lévy’s heuristic construction of Brownitn motion from its excursions has been made rigorous by other authors. We de ot attempt such a construction hers, nor do we give a complete spesiiatinn of the distribution of Brownian excursions, i the interest of intelligibility, : tion for the durations Content ourselves with the specification ofthe distri in rains Of the excursions. Sections 6.3 and 64 derive distributions fr fenctnfal® of Brownian motion involving its loc times we present in partite Feynman-Kac result for the so-alled “elastic” Brownian mat Us Oe Mulas of D. Williams and H. Taylor, and the Ray-Knie