————————————————————————————————— MAT4701: Voluntary Assignment 2 Sindre Froyn ————————————————————————————————— ——————————————– Problem 1 a) Let Xt and Yt be Ft -adapted and “reasonably bounded”. Prove that d Xt · Yt = Xt dYt + Yt dXt + dXt dYt ——————————————– Using g(t, x, y) = xy we can calculate the derivatives. ∂2g = 0, ∂x2 ∂g = y, ∂x ∂g = 0, ∂t ∂g = x, ∂y ∂2g = 0, ∂y 2 ∂2g =1 ∂y∂x ∂2g (t, Xt , Yt ) = 1 ∂y∂x ∂g ∂g (t, Xt , Yt ) = Yt , (t, Xt , Yt ) = Xt , ∂x ∂y Applying Itô’s multi dimensional formula. 1 d(Xt Yt ) = Yt dXt + Xt dYt + (dXt dYt + dYt dXt ) = Yt dXt + Xt dYt + dXt dYt 2 ——————————————– b) Let Yt be a martingale and compute the conditional expectation Z t Z t Yr dr | Fs = E E[Yr | Fs ]dr 0 0 when s ≤ t. Hint: split the integral in two cases. ——————————————– Z t Z s Z t E[Yr | Fs ]dr E[Yr | Fs ]dr + E[Yr | Fs ]dr = 0 0 s In the first term we now have Fs -measurability, and in the second term we use that Yr is a Martingale. Z s Z t Z s = Yr dr + Ys dr = Yr dr + Ys (t − s) 0 0 s ——————————————– Rt c) Let Yt be a martingale and put Xt = 0 Yr dBr . Compute d(Bt · Xt ) and find E[Bt · Xt | Fs ] when s ≤ t. ——————————————– We first note that Xt is an Itô process where Z t Yr dBr ⇐⇒ dXt = Yt dBt Xt = 0 1 Using the result from a), we know that d(Bt Xt ) = Xt dBt + Bt dXt + dBt dXt Substituting Yt dBt for dXt and using that dBt2 = dt. = Xt dBt + Bt Yt dBt + Yt dt = Yt dt + Xt + Bt Yt dBt Writing in integral form with B0 X0 = 0 since B0 = 0. Z t Z t (Xr + Br Yr )dBr Yr dr + Bt Xt = 0 0 Taking the conditional expectation. hZ t i hZ t i E[Bt Xt | Fs ] = E (Xr + Br Yr )dBr | Fs Yr dr | Fs + E 0 0 The first term was calculated in b), and the second term is a pure stochastic integral which means it is a martingale: Z s Z s (Xr + Br Yr )dBr Yr dr + Ys (t − s) + = 0 0 ——————————————– Problem 2 a) Let σ(t, ω) be any bounded Ft -adapted stochastic process. Solve the initial value problem dXt = −2tXt dt + σ(t, ω) X0 = 1 and compute E[Xt ]. Xt and Bt are 1-dimensional. ——————————————– 2 Introducing the integrating factor et . 2 2 2 d(et Xt ) = 2tet Xt dt + et dXt = 2 2 2tet Xt dt + et 2 − 2tXt dt + σ(t, ω) = ( ((2 (( 2 2 t ((( 2te Xt dt − 2tet Xt dt + et σ(t, ω) = et σ(t, ω) (( ( Writing in the integral form, and recalling that X0 = 1. Z t 2 t2 es σ(s, ω)dBs e Xt = X0 + 0 Z 2 Xt = X0 e−t + t es 2 −t2 σ(s, ω)dBs 0 −t2 Xt = e + Z t es 0 2 2 −t2 σ(s, ω)dBs Taking the expectation: h 2 E[Xt ] = E e−t ] + E Z t es 0 2 −t2 i 2 σ(t, ω)dBs = e−t 2 since e−t is a constant and the expectation to a stochastic integral is 0. ——————————————– b) Consider the 1-dimensional initial value problem dXt = b(t, Xt )dt + σ(t, Xt )dBt X0 = x Spell out in detail what conditions on b and σ are needed for a unique solution to exist. ——————————————– As per theorem 5.2.1 from the text book. |b(t, x) − b(t, y)| ≤ C|x − y| |σ(t, x) − σ(t, y)| ≤ C|x − y| |b(t, x)| ≤ C(1 + |x|) |σ(t, x)| ≤ C(1 + |x|) Both b and σ need to be Lipschitz continuous and they can have at most linear growth. The uniqueness is dependant on the Lipschitz property. Also E[|x|2 ] < ∞. ——————————————– In the rest if thus problem we will assume that b : [0, ∞) → R is a continuous function (note: no dependence on ω in b). We further assume that the coefficients in the initial value problem dXt = b(t) · Xt dt + σ(Xt )dBt X0 = x satisfy the conditions spelled out in b). c) Find Ex [Xt ]. ——————————————– We begin by solving the SDE, and do so by multiplying with R t the integrating −B(t) ′ factor e where B (t) = b(t), or equivalently B(t) = 0 b(s)ds. d e−B(t) Xt = −b(t)e−B(t) Xt dt + e−B(t) dXt = −b(t)e−B(t) Xt dt + e−B(t) b(t)Xt dt + σ(Xt )dBt = e−B(t) σ(Xt )dBt =⇒ Z t −B(t) e−B(s) σ(Xs )dBs e Xt = X0 + 0 Xt = xeB(t) + Z t eB(t)−B(s) σ(Xs )dBs 0 Taking the expectation, Ex [Xt ] = xeB(t) 3 ——————————————– d) Let h > 0 and assume that Xh (ω) is known. Compute Ex [Xt+h |Fh ]. ——————————————– By the Markov property, and using that f (x) = x, Ex [f (Xt+h ) | Fh ] = EXh [f (Xt )] = Ex [Xt ]x=Xh By the result in c), = Xh eB(t) ——————————————– Rt e) Prove that Yt = Xt · e− 0 b(s)ds is a martingale. ——————————————– Rt We’ll use B(t) = 0 b(s)ds, so Z t −B(t) e−B(s) σ(Xs )dBs . Yt = Xt e =x+ 0 There are three properties we must verify. First Yt must be Ft -measurable. This is trivial as we don’t use information from the future. Secondly Yt must be finite, which follows from the assumptions we have on σ(Xs ) given in this exercise. Thirdly E[Yt | Fu ] = Yu for u ≤ t. This point needs some special consideration. Z t i h e−B(s) σ(Xs )dBs | Fu = E[Yt | Fu ] = E x + 0 x+E hZ u −B(s) e i σ(Xs )dBs | Fu + E 0 hZ u t i e−B(s) σ(Xs )dBs | Fu = By measurability and independence, Z u i hZ t −B(s) e σ(Xs )dBs + E e−B(s) σ(Xs )dBs = x+ 0 u since expected stochastic integrals are 0, Z u e−B(s) σ(Xs )dBs = Yu . x+ 0 The three properties are satsified, and Yt is a martingale. ——————————————– Problem 3 Consider the initial value problem p X0 = x dXt = rXt dt + |Xt |dBt The coefficients in this equation do not satisfy the usual conditions for a unique solution to exist. Which condition fails? Prove your statement. ——————————————– 4 The stochastic part σ(Xt ) does not satisfy the Lipschitz property. To show this, we first see where the property is satisfied (as in equation 7.1.5 in the text book). σ(x) − σ(y) ≤ C|x − y| p p |x| − |y| ≤ C|x − y|. p p If we define a = |x| ≥ 0 and b = |y| ≥ 0, we get a2 = x ≥ 0 and b2 = y ≥ 0 |a − b| ≤ C|a2 − b2 | = C (a + b)(a − b) = C(a + b)|a − b| =⇒ 1 ≤ (a + b) C However, when we have a fixed C and choose small enough a and b such that 1 ≤ C(a + b) =⇒ 1 > (a + b) C the Lipschitz continuity is not satisfied, and we can’t guarantee uniqueness. ——————————————– Problem 4 Let B1 and B2 be two independent Brownian motions, and let τD be the first exit time of the diffusion dXt = (dB1 (t), dB2 (t)) from the unit ball D = {x = (x1 , x2 ) | x21 + x22 < 1}. Compute E x [τD ] when starting from a point inside D. Hint: Use the Dynkin’s formula on f (x1 , x2 ) = x21 + x22 . ——————————————– Dynkin’s formula states that, hZ τ i Ex [f (Xτ )] = f (x) + Ex Af (Xs )ds . 0 We begin by finding the generator Af (Xs ), which is given by Af (x) = X i bi (x) ∂2f 1X ∂f (σσ T )i,j (x) + . ∂xi 2 ∂xi ∂xj i,j Calculating the partial derivatives for i = 1, 2. ∂f = 2xi , ∂xi ∂2f = 2, ∂x2i ∂2f = 0 (i 6= j). ∂xi ∂xj From the diffusion dXt we see that bi = 0 and σi = 1 for i = 1, 2, so all the first derivatives are cancelled and σiT = σi . This results in Af (x) = 1 ∂2f 1 1X 2 σi (x) 2 = (1)(2) + (1)(2) = 2 2 2 2 ∂xi i 5 Obviously Af (Xs ) = 2. Using τ = σk = min(k, τD ) and the function f we have, when (x1 , x2 ) is the starting point in D, h Z σk i x 2 2 2 2 x E [B1 (σk ) + B2 (σk )] = x1 + x2 + 2E ds 0 1 1 − (x21 + x22 ) 2 When k → ∞ we get τD = lim σk < ∞ and Ex [σk ] ≤ Ex [τD ] = 1 − (x21 + x22 ) 2 ——————————————– Problem 5 Let dXt = 3dt + 2dBt , Xs = x, t ≥ s. a) Write down Lf (s, x) for the extended process Yt = (t, Xt ). ——————————————– By definition, for some f ∈ C 2 , LX f (x) = X bi (x) i ∂2f 1X ∂f (σσ T )i,j (x) + ∂xi 2 ∂xi ∂xj i,j 1 = b(x)f ′ (x) + σ 2 (x)f ′′ (x) = 3f ′ (x) + 2f ′′ (x). 2 For the extended process, LY f (s, x) = ∂f ∂2f ∂f +3 +2 2. ∂s ∂x ∂x ——————————————– b) Consider the optimal stopping problem Φ(s, x) = sup E(s,x) [e−2τ Xτ ]. τ ≥s Assume that this problem has an exit region D = {(s, x) | − ∞ < x < x0 } and use the verification theorem to determine x0 and Φ(s, x). ——————————————– By the verification theorem we have the requirement Lφ + f = Lφ = 0 (since f is 0 in this case). Lφ(s, x) = ∂φ ∂φ ∂2φ +3 +2 2 =0 ∂s ∂x ∂x 6 (from a) If we assume a solution on the form φ(s, x) = e−2s ψ(x) we get: Lφ(s, x) = −2e−2s ψ(x) + 3e−2s ψ ′ (x) + 2e−2s ψ ′′ (x) = 0 =⇒ −2ψ(x) + 3ψ ′ (x) + 2ψ ′′ (x) = 0 We have a standard second degree differential equation. Using the quadratic formula we find the roots r1 = 21 and r2 = −2 and get x ψ(x) = C1 e 2 + C2 e−2x . For x ∈ D we see that as x → −∞, C2 e−2x → ∞ when we require it to be equal to zero, so C2 = 0. Since we need to have continuity at the border of D, we have the additional requirement ψ(x0 ) = x0 . ψ(x0 ) = C1 e x0 2 = x0 C1 = x0 e− ⇒ x0 2 Since φ ∈ C 1 (G) in the verification theorem we can determine x0 . φ(s, x) = ∂φ(s, x) = ∂x x0 x e−2s x0 e− 2 e 2 e−2s x −2s e x0 − 2 e e−2s x0 2 x e2 if x ≤ x0 if x > x0 if x ≤ x0 if x > x0 Setting the derivatives equal in the point x = x0 , which is on the border of D these must be equal, and we can determine x0 . e−2s x0 − x 0 x 0 e 2 e 2 = e−2s 2 ⇒ x0 = 2 By the verification theorem x Φ(s, x) = 2e 2 −2s−1 . ——————————————– c) Solve the problem in b) again using dXt = 2dBt , Xs = x, t ≥ s. Prove that the exit region (the complement of D) is bigger than in b), and try to give an intuitive explanation for this. ——————————————– For this process the b functions are 0, so all the first derivatives disappear from the function. ∂2φ ∂φ +2 2 Lφ = ∂s ∂x Assuming φ(s, x) = e−2s ψ(x): −2e−2s ψ(x) + 2e−2s ψ ′′ (x) = 0 7 ⇒ ψ ′′ (x) − ψ(x) = 0 Finding the solutions for x2 − 1 = 0 which is r1 = 1 and r2 = −1 we have ψ(x) = C1 ex + C2 e−x = C1 ex since C2 = 0 because the second term diverges when x → −∞. By continuity we require ψ(x0 ) = C1 ex0 = x0 so C1 = x0 e−x0 . We have established that −2s e x0 e−x0 ex if x ≤ x0 φ(s, x) = e−2s x if x > x0 We differentiate and set these equal at the transitional point x0 so we can determine what it is. −2s ∂φ(s, x) e x0 e−x0 ex if x ≤ x0 = e−2s if x > x0 ∂x e−2s x0 e−x0 ex0 = e−2s ⇒ x0 = 1 By the verification theorem Φ(s, x) = ex−2s−1 . We have proved that the exit region with x0 = 1, which is [1, ∞), is bigger than the exit region from b) which is [2, ∞). The difference between the exit regions depends on the term 2dt which is only present in the process from b). This term provides us with a positive drift that counters the possible negative effects of the stochastic term. Because of this the process allows for a larger maximum stopping time. 8