Stochastic Analysis. Lecturer: Xue-Mei Li, Support Classes: Sebastian Völlmer 1 1 Problem Sheet 1, week 1 Let (Ω, F, P ) be a probability space. Exercise 1 Let X : Ω → W d be a bounded measurable map. Let µ = (X)∗ (P ) be the pushed forward measure on W d . Let πt : W d → Rd be the evaluation map, πt (ω) = ωt . Let µt = (πt )∗ µ. X Ω Wd X t =π t πt ◦X Rd Define the random variable Xt : Ω → Rd by Xt = πt ◦ X. Check that µt = (Xt )∗ (P ). Exercise 2 Let µ be the Wiener measure on (W0d , B(W0d )). Write πt = (πt1 , . . . , πtd ) in components. Then for Ai ∈ B(Rd ), t1 < t2 < · · · < tn , µ(πt1 ∈ A1 , . . . , πtk ∈ Ak ) Z = pt1 (0, y1 ) . . . pti+1 −ti (yi , yi+1 ) . . . ptk −tk−1 (yk−1 , yk ) Πnk=1 dyi . A1 ×···×Ak Answer the following questions: 1. Prove that Eπt = 0, E(πti )2 = t and E(πti πtj ) = 0 if i 6= j. 2. What is the probability distribution of πt ? 3. Prove that E(πsi πtj ) = (s ∧ t)δi,j . Here s ∧ t stands for min(s, t) and δi,j = 0 if i 6= j and δi,i = 1. 4. Define Gs = σ{πr : r ∈ [0, s]}. Let E{πt |Gs } be the conditional expectation of πt with respect to Gs . Show that E{πt |Gs } = πs . (If you are not familiar with conditional expectations, leave this problem until next week.) 5. Compute the probability distribution of |πt |2 . 6. Compute the distribution of πt − πs , for t > s. Stochastic Analysis. Lecturer: Xue-Mei Li, Support Classes: Sebastian Völlmer 2 7. Let f be a bounded measurable function. Define Pt f (x) = Ef (x + πt )). Show that for any f smooth with compact support, ∂Pt f 1 = ∆Pt f. ∂t 2 Exercise 3 Let (Bt ) be a real valued standard Brownian motion. Fom the definition of the Brownian compute the marginal distributions of (Bt ). Exercise 4 Let b : Rd → Rd be a Borel measurable functions. For each ω, consider the equation Z t xt (ω) = x0 + b(xs (ω))ds + Bt (ω). 0 Suppose that b is globally Lipschitz continuous, i.e. there is a number K > 0 s.t. |b(x) − b(y)| ≤ K|x − y|. Prove that for each initial value x0 there is a unique solution to the above equation. n g (B ), where g are Borel Hint to Problem 1(d). The class of functions πi=1 i si i measurable and 0 ≤ s0 < s1 < . . . sn = s < t, are sufficient for determining conditional expectation with respect to FsB . Stochastic Analysis. Lecturer: Xue-Mei Li, Support Classes: Sebastian Völlmer 3 Fix 0 ≤ s1 < s2 < · · · < sn ≤ s. For each i = 1, . . . , n, let gi : Ω → R be a bounded function measurable with respect to Gsi . Prove that n n E[f (πt − πs )πi=1 gi (πsi )] = Ef (πt − πs )E[πi=1 gi (πsi )]. Hint for Problem 3. Let n ∈ N , let 0 ≤ t1 < t2 < · · · < tn and f : Rn → R a bounded measurable function. Work on Ef (Bt1 , Bt2 , . . . Btn ). 4 Stochastic Analysis. Lecturer: Xue-Mei Li, Support Classes: Sebastian Völlmer 2 Problem Sheet 2, week 2 Let (Ω, F, P ) be a probability space. Let (Bt ) be a one dimensional Brownian motion unless otherwise stated. Exercise 5 1. Write down the Fourier transform of a Gaussian measure N (a, C). 2. Let a ∈ Rd and C = (Ckl ) a positive definite symmetric matrix. Let X = (X1 , . . . , Xd ) be an Rd valued random variable with probability distribution N (a, C). Prove that EX = a and cov(Xk , Xl ) = Ckl . 3. Let (Bt ) be an Rd valued Brownian motion. Let a ∈ Rd and A a d × dmatrix. For each t > 0, compute the probability distribution of ABt + at. Exercise 6 Let (Xt ) be a one dimensional process with finite dimensional distribution given below. For 0 = t0 < t1 < · · · < tk , Ak ∈ B(R), P (Xt1 ∈ A1 , . . . , Xtk ∈ Ak ) Z = pt1 (0, y1 ) . . . pti+1 −ti (yi , yi+1 ) . . . ptk −tk−1 (yk−1 , yk ) Πkk=1 dyi . A1 ×···×Ak Prove that (Xt ) is a Brownian motion. Exercise 7 (a) Let a 6= 0 be a real number. Show that motion; √1 Bat a is a Brownian (b) Let t0 ≥ 0. Prove that Bt0 +t − Bt0 is a standard Brownian motion; (c) Define a process Wt by W0 = 0 and Wt = tB 1 when t > 0. Show that Wt t is a Brownian motion. (d) Show that limt→∞ Bt t = 0. (e) Let Xt = Bt − tB1 , 0 ≤ t ≤ 1. Show that E(Xs Xt ) = s(1 − t) for s ≤ t. Compute the probability distribution of Xt . Exercise 8 A zero mean Gaussian process BtH is a fractional Brownian motion of Hurst parameter H, H ∈ (0, 1), if its covariance is 1 E(BtH BsH ) = (t2H + s2H − |t − s|2H ). 2 Then E|BtH −BsH |p = C|t−s|pH . If H = 1/2 this is Brownian motion (Otherwise this process is not even a semi-martingale). Show that BtH has Hölder continuous paths of order α < H. Stochastic Analysis. Lecturer: Xue-Mei Li, Support Classes: Sebastian Völlmer 3 5 Problem Sheet 3, week 3 Exercise 9 A family of real-valued functions (fα , α ∈ I), where I is an index set, is uniformly integrable (u.i.) if Z lim sup sup |fα |dµ = 0. C→∞ α {|fα |≥C} Let X : Ω → R be an integrable. Prove that the family of functions {E{X|G} : G is a sub σ-algebra of F} is uniformly integrable. Exercise 10 Let X : Ω → R be integrable. Let (Ft , t ≥ 0) be a family of σalgebras such that Fs ⊂ Ft for s ≤ t (a filtration). Define Xt = E{X|Ft }, t ≥ 0. Show that E{Xt |Fs } = Xs for any pair of numbers s, t with t > s ≥ 0. Exercise 11 Let Bt be a Brownian motion and for s ≥ 0 define Fs = σ(Br : 0 ≤ r ≤ t). Show that Bt is a Gt -Markov process, which means that for any bounded Borel measurable function f , E{f (Bt )|Gs } = E{f (Bt )|σ(Bs )}. Exercise 12 If Bt is a standard Brownian motion show that (a) For any s ≥ 0, Bt − Bs is independent of Fs where Fs = σ{Br : 0 ≤ r ≤ s}. (b) For any 0 ≤ s < t, E{(Bt − Bs )2 − (t − s)|Fs } = 0. (c) Let Mt denote one of the processes: Bt , Bt2 − t and exp(Bt − t/2). Show that martingale property holds: for all t ≥ s, E{Mt |Fs } = Ms . Discrete Time Martingales Exercise 13 Let X1 , X2 , . . . be independent random variables with EXi = 0. Let Fn = σ{X1 , . . . , Xn }. Let Sn = X1 + · · · + Xn . Show that for j = 1, 2, . . . , E{Sn+j |Fn } = Sn . Exercise 14 Let X1 , X2 , . . . be independent random variables with EXi = 1. Let Fn = σ{X1 , . . . , Xn }. Let Mn = Πnk=1 Xk . Show that for all n, j = 1, 2, . . . , E{Mn+j |Fn } = Mn . Stochastic Analysis. Lecturer: Xue-Mei Li, Support Classes: Sebastian Völlmer 6 Optional exercises The following exercises are optional. Exercise 15 Let Bt be a standard 1-dimensional Brownian motion. The Brownian bridge from 0 to 0 in time 1 is the Brownian motion conditioned to be at 0 at time 1. Let A, Ai ∈ B(R). Compute the distribution of Bt conditioned on B1 = 0, ı.e. P {Bt ∈ A|B1 }, and the conditional marginal distribution P {Bt1 ∈ A1 , . . . , Btn ∈ An |B1 = 0}. Exercise 16 Let X1 , X2 , . . . be independent random variables with EXi = 0. Let Fn = σ{X1 , . . . , Xn }. Let S0 = 0 and Sn = X1 + · · · + Xn . Let Cn = fn (X1 , . . . , Xn−1 ) for some Borel function fn : Rn → R. Define X I(C, X)n = Ck (Sk − Sk−1 ). 1≤k≤n This is called the martingale transform. Compute E{I(C, X)n −I(C, X)n−1 |Fn−1 }. Exercise 17 Let {Xn , n = 0, 1, 2 . . . } be an Fn -adapted stochastic process with Xn ∈ L1 and X0 = 0. Define Gn = E{Xn+1 − Xn |Fn }, n ≥ 1. P (a) Let A0 = A1 = 0 and An = n−1 j=1 Gj , n ≥ 2. Show that An ∈ Fn−1 (previsible). (b) Let M0 = 0 and Mn = Xn − An show that Mn is a martingale. Then Xn = Mn + An . This is the analogue of the Doob-Meyer decomposition for continuous time processes. (c) If Xn has another decomposition Xn = M̃n + Ãn , where {M̃n } is a martingale with M̃0 = 0, Ãn is process with Ãn ∈ Fn−1 and Ã0 = Ã1 = 0. Show that Mn = M̃n and An = Ãn a.s. If Xn is a sub-martingale show that An is an increasing process. (d) Let (Mn ) be a martingale with EMn2 < ∞ and M0 = 0 show that there is an increasing process An such that Mn2 = Nn + An where Nn is a martingale and An is an increasing process. Show that An − An−1 = E{(Mn − Mn−1 )2 |Fn−1 }. Stochastic Analysis. Lecturer: Xue-Mei Li, Support Classes: Sebastian Völlmer 4 7 Problem Sheet 4, week 4 (Martingales) Let (Ω, F, Ft ) be a filtered probability space. All processes here are real valued. Exercise 18 If Mt is an L2 bounded martingale (that is supt E(Mt )2 < ∞) show that for s < t, E(Mt − Ms )2 = EMt2 − EMs2 . Exercise 19 Let φ be a convex function s.t. φ(Xt ) ∈ L1 . Show that (a) If Xt is a sub-martingale and φ is increasing then φ(Xt ) is a sub-martingale. (b) If Xt is martingale then φ(Xt ) is a sub-martingale. Show that |Xt | is a submartingale. Exercise 20 Let (Ω, F, Ft , P ) be a filtered probability space. Let Q be a probability measure such that Q << P . Let f = dQ dP . Let Qt and Pt are restrictions of Q, P to Ft . Let ft = E{f |Ft }. Show that ft is a L1 bounded martingale and dQt dPt = ft . Exercise 21 Let (Bt ) be a 1-dimensional standard Brownian motion. Prove that (Bt2 − t) is a martingale. Exercise 22 Let (Bt ) be a 1-dimensional standard Brownian motion. Let 0 = t0 < t1 < · · · < tn . Let Fs = σ{Br : 0 ≤ r ≤ s}. Suppose that ai be real valued functions with ai measurable with respect to Fti . 1. Suppose that s ≥ ti+1 . Prove that E{ai (Bti+1 − Bti )|Fs } = ai (Bti+1 − Bti ). 2. Suppose that s < ti , prove that E{ai (Bti+1 − Bti )|Fs } = 0. 3. Suppose that ti ≤ s < ti+1 , prove that E{ai (Bti+1 − Bti )|Fs } = ai (Bs − Bti ). 4. Define Mt = n−1 X ai (Bti+1 ∧t − Bti ∧t ). i=0 Prove that Mt is a martingale. Here s ∧ t denotes min(s, t). Pn−1 5. Let ft be a stochastic process given by ft (ω) = i=0 ai (ω)1(ti ,ti+1 ] (t). Prove Itô’s isometry: Z t 2 E(Mt ) = E (fs )2 ds. 0 Stochastic Analysis. Lecturer: Xue-Mei Li, Support Classes: Sebastian Völlmer 5 8 Problem Sheet 5, week 5 (Stopping Times and Martingales) Let t ∈ [0, ∞) and let (Ω, F, Ft , P ) be a filtered probability space. All stochastic processes are assumed to be continuous unless otherwise stated. Exercise 23 Let (Xt ) be a continuous Ft adapted real valued process. Let (Ft ) be a right continuous filtration. Show that τ1 := inf{t > 0 : Xt > 1} is a stopping time. Exercise 24 Let (Mt ) be an integrable and Ft adapted continuous stochastic process. Then Mt is a martingale if EMT = EMS for any Ft stopping times T and S that takes at most two values. Exercise 25 Let (Xt : t ∈ I) be an integrable continuous adapted process. Suppose that for all bounded stopping times S ≤ T , EXT ≤ EXS prove that E{XT |FS } ≤ XS . Exercise 26 Let 0 < t0 < t1 < t1 < · · · < tn+1 . Let Hi ∈ Fti and H−1 ∈ Ft0 be real valued. Let n X Xt (ω) = H−1 (ω)1{0} (t) + Hi (ω)1(ti ,ti+1 ] (t). i=0 Prove that Xt is progressively measurable. Exercise 27 Prove that if (Xt , t ≥ 0) is a sub-martingale with supt E|Xt | < ∞, then limt→∞ Xt exists almost surely. Exercise 28 If (Xt , t ∈ [0, ∞)) is a right continuous martingale, prove that the following are equivalent. 1. limt→∞ Xt converges in L1 to a random variable X∞ . 2. There exists a L1 random variable XT s.t. Xt = E{XT |Ft }. 3. (Xt , t < ∞) is uniformly integrable. Exercise 29 Using the Optional stopping theorem to prove the following: (1) A positive local martingale Mt with M0 = 1 is a super-martingale. (2) A positive local martingale is a martingale if E|M0 | < ∞ and EMt = EM0 for all t > 0. Exercise 30 Let (Mt ) be a continuous local martingale. Prove that if Mt has finite total variation on [0, T ] then Mt = M0 , any t ≤ T . Stochastic Analysis. Lecturer: Xue-Mei Li, Support Classes: Sebastian Völlmer 9 Problem Sheet 6: The Quadratic Variation Process Exercise 31 Let (A1t ), (A2t ) be continuous processes of finite total variation on any interval [0, s]. Suppose that A10 = A20 = 0. Let (Mt ) and (Nt ) be local martingales. Show that if (Mt Nt − A1t ) and (Mt Nt − A2t ) are local martingales then A1t = A2t a.s. for all t. Exercise 32 Let (Bt , t ≥ 0) be a standard Brownian motion. For any u, the distributions of sups≤u Bs and |Bu | are equal. 1. Let T1 = inf{t > 0, Bt ≥ 1}. Prove that T1 < ∞ almost surely. 2. Prove that the family (Bt , 0 ≤ t ≤ 1) is uniformaly integrable. Is (Bt : t ≥ 0) uniformly integrable? Exercise 33 Let (Mt ) be a continuous local martingale with M0 = 0. Prove that (Mt ) is a martingale if supt≤T Mt ∈ L1 . Exercise 34 Let H02 be the space of L2 bounded sample continuous martingales vanishing at 0. We define a norm: kM k = E(M∞ )2 . Prove that H02 is a Hilbert space. Exercise 35 Let M ∈ H02 . Let M∞ ≡ limt→∞ Mt . Prove the following statements. (1) For all t > 0, E(Mt2 ) = EhM it . (2) Let τ be a stopping time prove that E(Mτ )2 = EhM iτ and (3) lim EhM it = EhM i∞ = sup E(Mt2 ) = E(M∞ )2 . t→∞ t<∞ Exercise 36 Let M n , M ∈ H02 . Show that the following statements are equivan − M )2 = 0; lent: limn→∞ supt≤∞ E(Mtn − Mt )2 = 0; limn→∞ E(M∞ ∞ n n limn→∞ EhM − M i∞ = 0; and limn→∞ E supt≤∞ (Mt − Mt )2 = 0. Exercise 37 Let M, N ∈ H02 . Let τ be a stopping time. Prove that and E(Mτ Nτ ) = EhM, N iτ and E(M∞ N∞ ) = EhM, N i∞ . Exercise 38 Let M, N be bounded continuous martingales with M0 = N0 = 0. If Mt and Nt are furthermore independent prove that their quadratic variation hM, N it vanishes. Stochastic Analysis. Lecturer: Xue-Mei Li, Support Classes: Sebastian Völlmer 10 Hints: Exercise 34) How do you prove the limit of a sequence of continuous functions is continuous? Exercise 35) We need to use uniform integrabiity, L2 convergence, and Doob’s inequality. Exercise 36) Recall Burkholder-Davis-Gundy inequality. Exercise 38) Define Gs = σ{Mr , Nr : 0 ≤ r ≤ s}. Is (Mt ) a (Gt ) -martingale? Stochastic Analysis. Lecturer: Xue-Mei Li, Support Classes: Sebastian Völlmer 11 Problem Sheet 7: Stochastic Integrals Exercise 39 Let Bt be 1-dimensional Brownian Motion, Rt Rt a) Show that hBt , 0 Bs3 dBs i = 0 Bs3 ds. Rt Rt b) Let fs , gs be continuous and adapted stochastic processes. Write h 0 fs dBs , 0 gs dBs i as an integral with respect to the Lebesque measure. R Rt 1 c) Prove that ( 0 (Bs )2 dBs ) is a martingale and compute E 0 (Bs )2 dBs . d) Write down theR semi-martingale decomposition of the stochastic process Rt s 0 Br d(Br + r) . 0 (2Bs + 1)d Exercise 40 Let M, N ∈ H02 and si < si+1 . 1. Let H be bounded and Fsi measurable. Let Xt = H(Msi+1 ∧t − Msi ∧t ); Yt = H(hM, N isi+1 ∧t − hM, N isi ∧t ). (a) Prove that (Xt ) is a martingale. (b) Verify that Yt = hM, N it by proving that (Xt Nt − Yt ) is a martingale. Rt (c) Let Ks = H1(si ,si+1 ] (t). Show that 0 Ks dMs = Xt by proving that Z hX, N it = t Ks dhM, N is . 0 2. Let Ks be an elementary process: Ks = h N X PN i=1 Ksi 1(si ,si+1 ] (t), Hi (M si+1 − M si ), N it = Z 0 i=1 Prove that t Ks dhM, N is . Exercise 41 Let {Bt , Wt1 , Wt2 } be independent Brownian motions. Let (xs ) be an adapted continuous stochastic process. 1. Prove that (Wt ) defined below is a Brownian motion, Z Wt = 0 t cos(xs )dWs1 Z + 0 t sin(xs )dWs2 . Stochastic Analysis. Lecturer: Xue-Mei Li, Support Classes: Sebastian Völlmer 12 2. Let sgn(x) = 1 if x > 0 and sgn(x) = −1 if x ≤ 0. Prove that (Wt ) is a Brownian motion if it satisfies the following relation Z t Wt = sgn(Ws )dBs . 0 3. Prove that the process (Xt , Yt ) is a Brownian motion if they satisfy the following relations, Z t Z t 1 1Xs ≤Ys dWs2 1Xs >Ys dWs + Xt = 0 0 Z t Z t Yt = 1Xs ≤Ys dWs1 + 1Xs >Ys dWs2 . 0 0 Stochastic Analysis. Lecturer: Xue-Mei Li, Support Classes: Sebastian Völlmer 13 Problem Sheet 8: Itô’s formula Exercise 42 Let Bt = (Bt1 , . . . , Btd ) be a BM show that hB i , B j it = δij t and n Z t X |Bt |2 = 2 Bsi dBsi + nt. i=1 0 (Hint: exercise 38.) Exercise 43 Prove that if H and K are continuous and adapted semi-martingales, Z t Z t hHK, Bit = Hr dhK, Bir + Kr dhH, Bir . 0 Rt Exercise 44 Interpret 0 0 s dBs by a Lebesque integral. Exercise 45 Let f, g : R → R be C 2 functions, Bt one dimensional Brownian motion. Write the following as an integral with respect to the Lebesque measure. a) hf (Bt ), g(Bt )i. b) h exp (M − 12 hM i), exp (N − 12 hN i)i where Mt and Nt are continuous local martingales. 1 Exercise 46 If Nt is a local martingale show that eNt − 2 hN,N it is a local martingale 1 and EeNt − 2 hN,N it ≤ 1. Exercise 47 Show that any positive local martingale Nt with N0 = 1 can be written in the form of Nt = exp (Mt − 12 hM, M it ) where Mt is a local martingale. Exercise 48 Let σ and b be smooth bounded functions from R to R. (a) Let Bt be a one dimensional Brownian motion. Let (xt , t ≥ 0) be a real valued adapted sample continuous process, s.t. for all t Z t Z t xt = x0 + σ(xs )dBs + b(xs )ds 0 Show that E(xt )2 0 < ∞. Rt Hint: Gronwall’s Lemma: If f (t) ≤ a(t) + 0 g(s)f (s)ds where a(t) is increasing, then Z t f (t) ≤ a(t) exp( g(s)ds). 0 (c) Can you modify the proof to show that E(supt≤T (xt )2 ) < ∞? 14 Stochastic Analysis. Lecturer: Xue-Mei Li, Support Classes: Sebastian Völlmer Problem Sheet 9: Stochastic Differential Equations Let Bt be a one dimensional Brownian motion on a given filtered probability space. Let σ, b : R → R be locally bounded and Borel measurable. Exercise 49 Give a solution to the one dimensional SDE: drt = dBt + n−1 rt dt. Exercise 50 Write down a solution to dxt = xt g(Bt )dBt where g : R → R is a bounded Borel measurable function. Verify you claim. Exercise 51 Let σ : R → R be BC 1 and f : R → R a solution of the ODE f˙ = σ(f ). For g : R → R locally bounded and Borel measurable, assume that dyt = dBt + g(f (yt ))dt has a solution yt . Let b = σg + 21 σ̇σ. Show that xt = f (yt ) solves dxt = σ(xt )dBt + b(xt )dt Exercise 52 Consider the SDE: dxt = σ(xt )dBt + b(xt )dt, where σ, b are continuous. Assume that σ > 0. Define the scale function − ṡ(x) = e 2b(y) 0 σ 2 (y) dy Rx . Then Ls = 0 and s is strictly increasing. Define ( = σ(s−1 (y))ṡ(s−1 (y)), if y is in the image of s σ̃(y) = = 0, if y is not in the image of s . Define yt = s(xt ). Show that yt solves: dyt = σ̃(yt )dBt . Prove that if b is bounded, then pathwise uniqueness holds for the SDE dxt = dBt + b(xt )dt. Exercise 53 Consider the SDE, with Stratonovitch integration, dxt = −yt ◦ dBt , dyt = xt ◦ dBt . Show that x2t + yt2 is independent of t. Exercise 54 Consider dxt = σ(xt )dBt + b(xt )dt. Assume that σ and b are locally Lipschitz continuous and are of at most linear growth: |σ(x)| ≤ C(1 + |x|), hx, b(x)i ≤ C(1 + |x|2 ). Let xt be a solution. Prove that the SDE is complete. 15 Stochastic Analysis. Lecturer: Xue-Mei Li, Support Classes: Sebastian Völlmer Problem Sheet 10: SDEs Exercise 55 Let σk = (σk1 , . . . , σkd ), b = (b1 , . . . , bd ) : Rd → Rd be smooth functions. Let ! d d m X 1 X X i j ∂ ∂2 L= + bl . σk σk 2 ∂xi ∂xj ∂xl i,j=1 l=1 k=1 Suppose that for C > 0, |σ(x)|2 ≤ c(1 + |x|2 ), hb(x), xi ≤ c(1 + |x|2 ). Let f (x) = |x|2 + 1. Prove that Lf ≤ af for some a. Exercise 56 Let B 1 and B 2 be independent Brownian motions. Compute the infinitesimal generator L and prove that the SDE does not explode: dxt = (x2t + yt2 )dBt1 dyt = (x2t + yt2 )dBt2 Hint: The SDE lives in R2 . The harmonic function V (x) = C log |x| + K can be used to build a Lyapunov function. Exercise 57 Let b : R → R be a bounded smooth function. Let ut be a C 2,1 bounded solution to 1 d2 ∂ ∂ut = +b ut ∂t 2 dx2 ∂x Prove that, Z ut (x) = Ef (x + Bt ) exp 0 t 1 b(x + Bs )dBs − 2 Z t 2 (b(x + Bs )) ds . 0