Scribd Logo
Upload

Welcome to Scribd!

  • 36 views

    Change of Measure For Brownian Motion

    Uploaded by

    Lecture 9
    Under a change of measure from P to an equivalent measure Q, a semimartingale X preserves its semimartingale structure. Specifically, if X has the decomposition X=X0 +M +A under P, where M is a locally square integrable martingale and A is a process of finite variation, then under Q we can write X=X0 + (M - A*) + (A + A*), where A* is an additional process of finite variation needed to make M - A* a locally square integrable martingale under Q. For a martingale M with continuous paths, we can take A* = p·V - V·p, where p is the Radon-Nikody

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd

    Change of Measure For Brownian Motion

    Uploaded by

    Lecture 9
    36 views4 pages
    Under a change of measure from P to an equivalent measure Q, a semimartingale X preserves its semimartingale structure. Specifically, if X has the decomposition X=X0 +M +A under P, where M is a locally square integrable martingale and A is a process of finite variation, then under Q we can write X=X0 + (M - A*) + (A + A*), where A* is an additional process of finite variation needed to make M - A* a locally square integrable martingale under Q. For a martingale M with continuous paths, we can take A* = p·V - V·p, where p is the Radon-Nikody

    Original Description:

    sotchastic project 10

    Original Title

    Project 10

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Under a change of measure from P to an equivalent measure Q, a semimartingale X preserves its semimartingale structure. Specifically, if X has the decomposition X=X0 +M +A under P, where M is a locally square integrable martingale and A is a process of finite variation, then under Q we can write X=X0 + (M - A*) + (A + A*), where A* is an additional process of finite variation needed to make M - A* a locally square integrable martingale under Q. For a martingale M with continuous paths, we can take A* = p·V - V·p, where p is the Radon-Nikody

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    36 views4 pages

    Change of Measure For Brownian Motion

    Uploaded by

    Lecture 9
    Under a change of measure from P to an equivalent measure Q, a semimartingale X preserves its semimartingale structure. Specifically, if X has the decomposition X=X0 +M +A under P, where M is a locally square integrable martingale and A is a process of finite variation, then under Q we can write X=X0 + (M - A*) + (A + A*), where A* is an additional process of finite variation needed to make M - A* a locally square integrable martingale under Q. For a martingale M with continuous paths, we can take A* = p·V - V·p, where p is the Radon-Nikody

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    of 4

    P10-1

    Project 10

    1.

    Change of measure for Brownian motion


    Let {Bt : 0 t 1} be a Brownian motion with respect to a (standard)
    ltration {Ft } on (, F, P). Write U for its quadratic variation process, Ut = t.
    For each R, the process
    
    
    qt = exp Bt 12 2 t
    for 0 t 1
    is a nonnegative martingale, with Pqt = Pq0 = 1. Dene a new probability
    measure Q on F1 by specifying q1 to be its density with respect to P. That is,
    Q X = P(Xq1 )
    at least for all bounded random variables X .
    Show that Q is equivalent to P, in the sense that both measures have the
    same collection N of negligible sets.
    Show that Q X = P(Xqt ) if X is Ft -measurable. Explain why qt is a
    Radon-Nikodym density for Q with respect to P when both measures are
    restricted to Ft .
    For xed s and t = s + , a xed F in Fs , and a bounded measurable f ,
    show that
    
    
    
    Q F f (Bt Bs ) = P(Fqs ) P f (Bt Bs ) exp (Bt Bs ) 12 2
    
    
    
    1
    = Q F
    f (z) exp 12 (z )2 / dz

    Deduce that, under Q , the process Bt t is a standard Brownian motion.

    2.

    The Black-Scholes formula


    Stock prices (in units so that S0 1) are sometimes modeled by a continuous
    process driven by a Brownian motion, B, on [0, 1];
    St = exp(( 1/2 2 )t + Bt )
    for 0 t 1
    2
    
    1
    = exp( Bt /2 t)
    where 
    Bt = Bt + (/ )t
    for constants > 0 (assumed known) and (unknown). That is,
    St = (Bt , Ut )

    I am ignoring ination. cf.


    expression of value of stock as
    a multiple of a bond price.

    where (x, y) = exp( x + ( 1/2 2 )y).

    Suppose Y = f (S), with f a C-measurable functional on C[0, 1]. How much


    should one pay at time 0 in order to receive the amount Y at time 1?
    Use the Ito formula to show that
    <1>

    St = 1 + S Bt + S Ut ,

    In more traditional notation,


    d St
    = d Bt + dt.
    St
    Roughly speaking, the relative increments of S behave like the increments of
    a Brownian motion with drift . The process S B is the locM20 [0, 1] part
    of the semimartingale decomposition of S.
    d St = St d Bt + St dt,

    Statistics 603a: 11 November 2004

    or

    c David Pollard
    

    P10-2

    Similarly, show that St = 1 + S 


    Bt .
    Show that Y can be written as a C-measurable functional of the 
    B sample
    path.
    Temporarily suppose that = 0, so that 
    B is a standard Brownian motion.
    (i) Use stochastic calculus to show that
    1
    
    B=
    S
    S
    Hint: What do you know about the increments of the process that takes
    a constant value?
    (ii) Suppose PY 2 < . Invoke results from Project 9 to show that there
    exists a predictable H such that
    H
    <2>
    Y = PY + H 
    B1 = PY + K S1
    where K :=
    .
    S
    (iii) Interpret the last equality as an assertion that there exists an (idealized?)
    hedging stategy that returns Y PY . Deduce that the arbitrage price
    for Y equals PY in the special case where = 0.
    Now consider the case where is unknown, possibly nonzero. Let Q be
    the probability measure with density exp( B1 12 2 ) with respect to P,
    where = / . Show that 
    B is a standard Brownian motion under Q .
    2
    Assume that Q Y < . Show that there exists some predictable
    process K (in some apppropriate L2 space) for which
    Y = Q Y + K S1

    almost surely [Q ].

    I believe that the threat of the trading scheme that delivers a return K S1
    now forces Q Y to be the amount one should pay at time 0 to receive the
    amount Y at time 1. What do you think? Should the fact that K seems
    to depend on the unknown invalidate the arbitrage argument?
    Suppose Y actually depends only on the stock price at time 1, that is,
    Y = f 1 (S1 ) for some measurable function f 1 . Show that
    
    
    where W N (0, 1) under Q.
    Q Y = Q f 1 exp( W 12 2 )
    Deduce that Q Y does not depend on .
    Specialize even further, to the case where f 1 (x) = (x C)+ , for some
    constant C, to derive the famous Black-Scholes formula for the price of a
    European option.

    3.

    I should reread Harrison &


    Pliska (1981).

    Does K actually depend on ?


    As I type this Project late at night, I nd myself in the embarrassing position of
    not really understanding how the question is handled for a general Y . However,
    when Y = f 1 (S1 ) there is another approach that avoids the difculty by
    constructing an explicit strategy via the solution to a partial differental equation.
    Look for a smooth function f (x, t) for which f (x, 1) = f 1 (x) and
    2 x 2 f x x (x, t) + f t (x, t) = 0
    Use Ito to show that
    f (St , t) = f (1, 0) + Fx St

    almost surely [P].

    Show that, under Q , the stock price process is a martingale. Deduce that
    
    
    f (St , t) = Q f (S1 , 1) | Ft = Q (Y | Ft )
    Statistics 603a: 11 November 2004

    c David Pollard
    

    P10-3

    and, in particular, f (1, 0) = Q (Y | F0 ) = Q Y .


    I hope I can sort through my confusion before the lecture. I will reread
    the nal section of Chung & Williams (1990).

    4.

    Change of measure for semimartingales


    The key fact about the change from P to an equivalent measure Q is the
    preservation of the semimartingale property. It is not at all an obvious fact.
    For suppose that X is a P-semimartingale that has decomposition X 0 + M + A,
    where M is a locally square integrable P-martingale and A FV0 . Under Q the
    A process is still in FV0 , but we will have to subtract another FV0 process A
    from M to make it a locally square integrable Q-martingale, leading to the
    Q-semimartingale decomposition X = X 0 + (M A ) + (A + A ).
    To establish these facts in the general case I would need some theory about
    processes with jumpsthings like the Doob-Meyer decomposition. Using only
    tools developed in the course, I can show you how to treat a special case.
    Consider only a process M locM20 ([0, 1], P) with continuous sample paths.
    Here I have added the P to emphasize that the martingale properties hold
    under the P distribution. Suppose that P and Q are equivalent measures, with
    q1 := dQ/dP and dP/dQ = p1 = 1/q1 . Assume that the cadlag versions
    of the P-martingale qt := P(q1 | Ft ) and the Q-martingale pt := Q( p1 | Ft )
    actually have continuous sample paths.
    Show that p = Q( p1 | F ) for each [0, 1]-valued stopping time .
    Explain why we can assume pt qt 1. More specically, explain why
    pt can be thought of as the density of P with respect to Q when both
    measures are restricted to Ft .

    Dene
    k := 1 inf{t : pt k or pt 1/k} inf{t : |Mt | k}.
    Without loss of generality, we may also assume that Mk M20 ([0, 1], P) .
    Show that pM locM20 ([0, 1], Q). Hint: For s < t and F Fs show that
    
    
    
    
    QF ptk Mtk psk Msk = QF ptk Mtk psk Msk {k > s}

    Argue that F{k > s} Fsk then deduce


     that the right-hand side of the
    last equality equals PF Mtk Msk {k > s} = 0.
    Use the fact that q and M are both in locM20 ([0, 1], P) to explain
    why the process Y := q M V , where V := [q, M] FV0 , is also
    in locM20 ([0, 1], P). Hint: First explain why Ytk = q Mtk + M qtk .
    Explain why both Y and V have continuous sample paths.
    Explain why pY locM20 ([0, 1], Q).
    Explain why [ p, V ] 0. Hint: V is a FV0 process with continuous
    sample paths.
    Deduce that pt Vt = p Vt + V pt .
    Deduce that M p V V p locM20 ([0, 1], Q).

    Statistics 603a: 11 November 2004

    c David Pollard
    

    P10-4

    Explain why V p locM20 ([0, 1], Q). Hint: p is a Q-martingale.


    Explain why A := p V is in FV0 .
    Conclude that M A locM20 ([0, 1], Q).
    You should check that this recipe works for the Brownian motion example
    in Section 1.

    5.

    Things I could show if I had more time


    (Actually I would also need some facts about processes with jumps.)
    (i) Every local-martingale is a semimartingale.
    (ii) Suppose P and Q are equivalent probability measures. If X is a
    P-semimartingale then it is also a Q-semimartingale. Moreover, for
    H locHBdd , the stochastic integral H X when calculated using the
    methods from Project 7 under P is the same as the stochastic integral
    when calculated under Q. The last assertion can be proved using the
    characterization of the stochastic integral given in Project 7.
    References
    Chung, K. L. & Williams, R. J. (1990), Introduction to Stochastic Integration,
    Birkhauser, Boston.
    Harrison, J. M. & Pliska, S. R. (1981), Martingales and stochastic integrals
    in the theory of continuous trading, Stochastic Processes and their
    Applications 11, 215260.

    Statistics 603a: 11 November 2004

    c David Pollard
    

    You might also like