Chap 11. Introduction to Jump Process Stochastic Calculus for Finance II Steven E. Shreve 財研二 范育誠 AGENDA 11.5 Stochastic Calculus for Jump Process 11.5.1 Ito-Doeblin Formula for One Jump Process 11.5.2 Ito-Doeblin Formula for Multiple Jump Process 11.6 Change of Measure 11.7 Pricing a European Call in Jump Model Ito-Doeblin Formula for Continuous-Path Process For a continuous-path process, the Ito-Doeblin formula is the following. Let X c t X 0 0 s dW s 0 s ds c t t In differential notation, we write dX c t s dW s s ds Let f x C 2 where C 2 : 1st and 2nd derivatives are defined and continuous. Then df X c s f X c s dX c s 1 f X c s dX c s dX c s 2 1 f X c s s dW s f X c s s ds f X c s 2 s ds 2 Write in integral form as f X c t f X c 0 f X c s s dW s f X c s s ds t t 0 0 1 t f X c s 2 s ds 2 0 Ito-Doeblin Formula for One Jump Process Add a right-continuous pure jump term J X t X 0 I t R t J t where Define I t : Ito integral term R t : Riemann integral term J t : Pure jump term X c t X 0 I t R t Between jumps of J 1 df X s f X s dX s f X s dX s dX s 2 f X s s dW s f X s s ds f X s dX c s 1 f X s 2 s ds 2 1 f X s dX c s dX c s 2 Ito-Doeblin Formula for One Jump Process Theorem 11.5.1 Let X t be a jump process and f x C 2 . Then f X t f X 0 f X s dX c s t 0 f X s f X s 1 t c c f X s dX s dX s 2 0 0 s t PROOF : Fix , which fixes the path of X , and let 0 1 2 n1 t be the jump times in 0,t of this path of the process X . We set 0 0 is not a jump time, and n t, which may or may not be a jump time. PROOF (con.) Whenever u v , u, v j , j 1 f X v f X u f X c s dX c s v u 1 v c c f X s dX s dX s 2 u Letting u j , v j 1 and using the right-continuity of We conclude that f X f X j 1 j 1 j j f X s dX c s 1 j1 f X s dX c s dX c s 2 j Now add the jump in f x at time j 1 f X j 1 f X j j 1 j 1 j1 f X s dX s f X s dX c s dX c s 2 j c f X j 1 f X j 1 X PROOF (con.) Summing over j 0,1, , n 1 f X t f X 0 n 1 f X j 1 f X j j 0 t 0 1 t f X s dX s f X s dX c s dX c s 2 0 c n 1 f X j 1 f X j 1 j 0 Example (Geometric Poisson Process) Geometric Poisson Process S t S 0 exp N t log 1 t S 0 e t 1 We may write S t S 0 f X t f x ex where X t t N t log 1 0 t S 0 1 S 0 0u t t S 0 0 S u du S u S u S u If there is no jump at time u 1 t f X u dX c u dX c u f X u f X u 2 0 0u t S u 1 du S u S u 0 S 0 S 0 0u t t 1 1 S u du S u N u 0 1 S u S u 1 We have S u S u S u N u S t f X t S 0 t If there is a jump at time u S u S u 0 X c t t f X 0 f X u dX c u N t 1 t S u dN u S 0 0 Example (con.) S t S 0 S u du S u dN u t t 0 0 M u N u u t S 0 S u dM u 0 In this case, the Ito-Doeblin formula has a differential form dS t S t dM t S t dt S t dN t Independence Property Corollary 11.5.3 W t : Brownian motion N t : Poisson process , 0 defined on the same space , F, and relative to the same filtration F t W t and N t are independent PROOF : 1 Y t exp u1W t u2 N t u12t eu2 1 t 2 1 Define f x e x , X s u1W s u2 N s u12 s eu2 1 s 2 1 X c s u1W s u12 s eu2 1 s 2 1 dX c s u1dW s u12 ds eu2 1 ds 2 dX c s dX c s u12 ds PROOF (con.) If Y has a jump at time s, then 1 Y s exp u1W s u2 N s 1 u12 s eu2 1 s Y s eu2 2 u2 Y s Y s e 1 Y s Therefore, Y s Y s eu2 1 Y s N s According to Ito-Doeblin formula Y t f X t f X 0 f X s dX c s 1 t f X s dX c s dX c s f X s f X s 0 2 0 0 s t t t t t 1 1 1 u1 Y s dW s u12 Y s ds eu2 1 Y s ds u12 Y s ds Y s Y s 0 0 0 0 2 2 0 s t t 1 u1 Y s dW s eu2 1 Y s ds eu2 1 Y s dN s t t t 0 0 0 1 u1 Y s dW s eu2 1 Y s dM s t t 0 0 PROOF (con) Y is a martingale and Y 0 1, EY t 1 for all t. In other words 1 exp u1W t u2 N t u12t eu2 1 t 1 2 e u1W t u2 N t e e 1 2 u1 t eu2 1 t 2 The corollary asserts more than the independence between N(t) and W(t) for fixed time t, saying that the process N and W are independent. For example, max0st W s is t independent of 0 N s ds AGENDA 11.5 Stochastic Calculus for Jump Process 11.5.1 Ito-Doeblin Formula for One Jump Process 11.5.2 Ito-Doeblin Formula for Multiple Jump Process 11.6 Change of Measure 11.7 Pricing a European Call in Jump Model Ito-Doeblin Formula for Multiple Jump Process Theorem 11.5.4 (Two-dimensional Ito-Doeblin formula) f C 2 , X 1 , X 2 are jump processes Continuous Part Jump Part Ito’s Product Rule for Jump Process Corollary 11.5.5 PROOF : f x1 , x2 x1 x2 f x1 x2 , f x2 x1 , f x1 x1 f x2 x2 0 , f x1 x2 f x2 x1 1 cross variation PROOF (con.) X 1 0 X 2 0 X 2 s dX 1c s X 1 s dX 2c s X 1c , X 2c t t t 0 0 X s X s X s X s 0 s t 1 2 1 2 X1 t X1c t J1 t , X 2 t X 2c t J 2 t are pure jump parts of X1 t and X 2 t , respectively. PROOF (con.) Show the last sum in the previous slide is the same as X s X s X s X s 0 s t 1 2 1 2 Doleans-Dade Exponential Corollary 11.5.6 Let X t be a jump process. The Doleans-Dade exponential of X is defined to be the process 1 Z X t exp X c t X c , X c t 1 X s 2 0 s t It is the solution to the S.D.E. dZ X Z X t dX t with Z X 0 1 integral form is Z X t 1 Z X s dX s t 0 Comparison (Girsanov ' s THM ) : 1 t t Z t exp s dW s 2 s ds 2 0 0 1 exp X c s X c , X c s 2 dZ t Z t dX c t PROOF PROOF of Corollary 11.5.6 : X t X c t J t where X c t s dW s s ds t t 0 0 1 Define Y t exp X c t X c , X c t 2 t 1 t 2 t exp s dW s s ds s ds 0 2 0 0 From the Ito-Doeblin formula dY t Y t dX c t Y t dX c t Note : We define K(0)=1 1 X s K t K t 1 X t K t K t K t K t X t Define K t 0 s t PROOF (con.) Use Ito’s product rule for jump process to obtain [Y,K](t)=0 dZ X Z X t dX t with Z X 0 1 AGENDA 11.5 Stochastic Calculus for Jump Process 11.6 Change of Measure 11.6.1 Change of Measure for a Poisson Process 11.6.2 Change of Measure for a Compound Poisson Process 11.6.3 Change of Measure for a Compound Poisson Process and a Brownian Motion 11.7 Pricing a European Call in Jump Model Change of Measure for a Poisson Process t Z t e Define N t Lemma 11.6.1 Z t dM t Z t is a martingale under P and Z t 1 for all t. dZ t PROOF : M t M(t)=N(t)-λt X c t t , J t N t Define X t Then X c , X c t 0, and if there is a jump at time t, then X t so 1 X t PROOF (con.) Z(t) may be written as 1 Z t exp X c t X c , X c t 1 X s 2 0 s t We can get the result by corollary 11.5.6 Let X t be a jump process. The Doleans-Dade exponential of X is defined to be the process 1 Z X t exp X c t X c , X c t 1 X s 2 0 s t It is the solution to the S.D.E. dZ X Z X t dX t with Z X 0 1 integral form is Z X t 1 Z X s dX s t 0 Change of Poisson Intensity Theorem 11.6.2 Under the probability measure P , the process N t , 0 t T is Poisson with intensity PROOF : Example (Geometric Poisson Process) Geometric Poisson Process S t S 0 exp t N t log 1 t 1 t e S t is a martingale under P-measure, and hence S t has mean rate of return α. dS t S t dt S t dM t 11.6.4 M t N t t, N t is a Poisson process with intensity under P We would like to change measure such that dS t rS t dt S t dM t 11.6.5 M t N t t, N t is a Poisson process with intensity under P S 0 e t N t Example (con.) The “dt” term in (11.6.4) is S t dt 11.6.6 The “dt” term in (11.6.5) is 11.6.7 r S t dt Set (11.6.6) and (11.6.7) are equal, we can obtain r We then change to the risk-neutral measure by To make the change of measure, we must have 0 r t Z t e N t If the inequality doesn’t hold, then there must be an arbitrage. Example (con.) 0 r If 0 , then S t S 0 ert 1 N t S 0 ert Borrowing money S(0) at the interest rate r to invest in the stock is an arbitrage. AGENDA 11.5 Stochastic Calculus for Jump Process 11.6 Change of Measure 11.6.1 Change of Measure for a Poisson Process 11.6.2 Change of Measure for a Compound Poisson Process 11.6.3 Change of Measure for a Compound Poisson Process and a Brownian Motion 11.7 Pricing a European Call in Jump Model Definition N t is a Poisson process with intensity Y1 , Y2 ,... are i.i.d. random variables defined on a probability space ,F,P Assume that Yi is independent of N t Compound Poisson Process N t Q t Yi i 1 If N jumps at time t, then Q jumps at time t and Q t YN t Jump-Size R.V. have a Discrete Distribution Yi takes one of finitely many possible nonzero values y1 , y2 , ... , yM p ym P Yi ym , m 1, 2, ... ,M M p y 1 m 1 m According to Corollary 11.3.4 M N t Nm t m 1 Nm t is the number of jumps in Q t of size ym up to and including time t N1 , N2 , ... ,NM are independent and each Nm has intensity m p ym N t M i 1 m 1 Q t Yi ym N m t Jump-Size R.V. have a Discrete Distribution Let 1 , 2 , ... ,M be given positive numbers, and set Lemma 11.6.4 The process Z t is a martingale. In particular, EZ t 1 for all t. PROOF of Lemma 11.6.4 From Lemma 11.6.1, we have Z m is a martingale. For m n, Nm and Nn have no simultaneous jumps Zm , Zn 0 By Ito’s product rule Z1 , Z 2 are martingales and the integrands are left-continuous Z1 Z 2 is a martingale In the same way, we can conclude that Z t Z1 t Z2 t Z M t is a martingale. Jump-Size R.V. have a Discrete Distribution Because Z T 0 almost surely and EZ T 1 , we can use Z T to change the measure, defining P A Z T dP A for all Z F Theorem 11.6.5 (Change of compound Poisson intensity and jump distribution for finitely many jump sizes) Under P , Q Mt is a compound Poisson process with intensity m , and Y1 , Y2 , are i.i.d. R.V. with m 1 m P Yi ym p ym Zm t e m m t m m M Z t Zm t m 1 Nm t PROOF of Theorem 11.6.5 Jump-Size R.V. have a Continuous Distribution The Radon-Nykodym derivative process Z(t) may be written as We could change the measure so that Q t has intensity and have a different density f y by using the M Radon-Nykodym derivative process Y1 , Y2 ,... N t x m 1 m Notice that, we assume that f y 0 whenever f y 0 m N t Xi i 1 AGENDA 11.5 Stochastic Calculus for Jump Process 11.6 Change of Measure 11.6.1 Change of Measure for a Poisson Process 11.6.2 Change of Measure for a Compound Poisson Process 11.6.3 Change of Measure for a Compound Poisson Process and a Brownian Motion 11.7 Pricing a European Call in Jump Model Definition Compound Poisson Process Let 0 , f y 0 whenever f y 0 , t is an adaptive process Lemma 11.6.8 The process Z t of (11.6.33) is a martingale. In particular, EZ t 1 for all t 0 PROOF : Z1 t is continuous Z1 , Z 2 t 0 Z 2 t has no Ito integral part By Ito’s product rule, Z1(s-),Z2(s-) are left-continuous Z1(s),Z 2(s) are martingales Z1(t)Z2(t) is a martingale Theorem 11.6.9 Under the probability measure P , the process is a Brownian motion, Q t is a compound Poisson process with intensity and i.i.d. jump sizes having density f y , and the processes W t and Q t are independent. PROOF : The key step in the proof is to show Y u euy f y dy ? Is Θ independent with Q (or Z2) ? PROOF (con.) Define We want to show that X1 t Z1 t , X 2 t Z2 t , X1 t Z1 t X 2 t Z2 t are martingales under P. No drift term. So X1(t)Z1(t) is a martingale PROOF (con.) The proof of theorem 11.6.7 showed that X2(t)Z2(t) is a martingale. Finally, because X1(t)Z1(t) is continuous and X2(t)Z2(t) has no Ito integral part, [X1Z1,X2Z2](t)=0. Therefore, Ito’s product rule implies X2(s)Z2(s), X1(s)Z1(s) are martingales. X1(s-)Z1(s-), X2(s-)Z2(s-) are left-continuous. Theorem 11.4.5 implies that X1(t)Z1(t)X2(t)Z2(t) is a martingale. It follows that Theorem 11.6.10 (Discrete type) Under the probability measure P , the process is a Brownian motion, Q t is a compound Poisson process with intensity and i.i.d. jump sizes satisfying PYi ym p ym for all i and m 1, 2,..., M , and the processes W t and Q t are independent. AGENDA 11.5 Stochastic Calculus for Jump Process 11.6 Change of Measure 11.7 Pricing a European Call in Jump Model 11.7.2 Asset Driven by Brownian Motion and Compound Poisson Process Definition Q(t)-λβt is a martingale. Theorem 11.7.3 The solution to is PROOF of Theorem 11.7.3 Let We show that is a solution to the SDE. X is continuous and J is a pure jump process → [ X,J ](t)=0 PROOF (con.) The equation in differential form is