STK4510 - Eksamen 2009 ————————————————————————————————— Exercise 1 (a) Show that Brownian motion Bs is Ito integrable on any interval [0, t]. ————————————————————————————————— Definition 3.1.4 (Oksendal): Ito integrability. A function f is Ito integrable, denoted f ∈ V = V[S, T ] if the following three properties are satisfied: (i) (t, ω) 7→ f (t, ω) is B × F -measurable. (ii) f (t, ω) is Ft -adapted. Z T (iii) E[ f (t, ω)2dt] < ∞. S We have to verify these three properties for f (s, ω) = Bs (ω). Properties (i) and (ii) are obvious for Brownian motion. Brownian motion Bt is Ft -adapted and B × F -measurable. For property (iii), finite variance, we calculate the result: Z t Z t Z t t2 2 2 E[ Bs ds] = E[Bs ]ds = sds = < ∞. 2 0 0 0 The three properties are verified for BM, so Bs is Ito integrable for any interval [0, t]. ————————————————————————————————— (b) Rt Apply the Ito Formula to calculate 0 Bs dBs . Find the expectation and variance of the integral. ————————————————————————————————— We apply Ito’s formula to the function g(t, x) = x2 where x = Bs . Finding the partial derivatives: ∂g = 0, ∂t ∂g = 2x, ∂x ∂2g = 2. ∂x2 ∂g ∂g 1 ∂2g (t, Bt )dt + (t, Bt )dBt + (t, Bt )(dBt )2 d g(t, Bt ) = ∂t ∂x 2 ∂x2 1 = 0 + 2Bt dBt + (2)dt 2 = 2Bt dBt + dt 1 Writing out in integral form, and exchanging g(t, Bt ) = Bt2 : Z t Z t 2 2 Bt = B0 + 2 Bs dBs + ds 0 BM starts in 0, so B02 = 0 and Bt2 Rt 0 0 ds = t. =2 Z t Bs dBs + t 0 Shifting terms and dividing by 2: Z t 1 1 Bs dBs = Bt2 − t. 2 2 0 Finding the expectation: Z t E[ Bs dBs ] = 0 0 since the expectation to stochastic integrals always is 0. Alternatively, 1 1 1 1 1 1 E[ Bt2 − t] = E[Bt2 ] − t = t − t = 0. 2 2 2 2 2 2 For the variance, denoting the integral with I: "Z Var[I] = E[I 2 ] − E[I]2 = E | {z } =0 Ito isometry: = Z t E[Bs2 ]ds = 0 Z 0 2 t t Bs dBs 0 t2 sds = . 2 2 # ————————————————————————————————— (c) Use directly the definition of Ito integration to show that Z t Z t sdBs = tBt − Bs ds. 0 0 ————————————————————————————————— Definition 3.1.6 (Oksendal): The Ito Integral. For f ∈ V[0, t] the Ito integral of f from 0 to t is defined by: Z t Z t f (s, ω)dBs = lim φn (s, ω)dBs(ω) n→∞ 0 0 where {φn } is a sequence of elementary functions such that Z t 2 E f (s, ω) − φn (s, ω) dt → 0 as n → ∞. 0 We choose a partition of [0, t] with 0 = s0 < s1 < . . . < sN = t. We define the elementary process: φn (s) = N −1 X j=0 sj 1(sj ≤ s < sj+1). We must show that φn → s, which means the following expectation must converge to 0: Z t 2 E φn (s) − s ds 0 Inserting the definition of the elementary process: "Z # −1 2 tN X E sj 1(sj ≤ s < sj+1 ) − s ds 0 j=0 — Unknown steps — N −1 Z sj+1 X j=0 — Unknown steps — sj (sj − s)2 ds N −1 1X (sj+1 − sj )3 3 j=0 3 Telescoping sum. = n−1 X 1 max (sj+1 − sj ) (sj+1 − sj )2 ≤ 3 0≤j≤N −1 j=0 1 max(sj+1 − sj )t2 −→ 0 as n → ∞ or ∆sj → 0 3 j Equality verified: Z t sdBs = 0 Z n−1 X t φn (s)dBs = 0 j=0 (exp 2?) (exp 2?) sj (Bsj+1 − Bsj ). Define Bj = Bsj . tBt = N −1 X j=0 = N −1 X j=0 = N −1 X j=0 = N −1 X j=0 = N −1 X j=0 sj+1 Bj+1 − sj Bj sj+1 Bj+1 −sj Bj+1 + sj Bj+1 − sj Bj sj+1 Bj+1 − sj Bj+1 + Bj+1 (sj+1 − sj ) + Bj+1 (sj+1 − sj ) + N −1 X j=0 N −1 X j=0 Z sj Bj+1 − sj Bj sj (Bj+1 − Bj ) t sdBs 0 Recall that the Ito integral needs the sample point must the the leftmost point in the interval. For deterministic integrals, (Riemann/Stieltjes) the sample point can be arbitrary. Remains to verify the last integral: the following expectation must converge to 0. !2 Z t N −1 X E Bj+1(sj+1 − sj ) − Bs ds 0 j=0 Note that Bj+1 (sj+1 − sj ) = calculus. N −1 Z X E j=0 R sj+1 Bj+1ds by the fundamental theorem of !2 Z t sj+1 Bj+1 ds − Bs ds sj 0 sj 4 We can partition the second integral into N − 1 smaller integrals, and collect them together. !2 N −1 Z sj+1 X E Bj+1 − Bs ds j=0 sj Split the indices: N −1 X "Z E i,j=0 si+1 (Bi+1 − Bs )ds si Z sj+1 sj (Bj+1 − Bs )ds # Consider two possibilities: i = j and i 6= j, by symmetry we can collect the i > j and j > as two times one of the cases. We assume j > i. !2 Z sj+1 N −1 Z si+1 Z sj+1 N −1 X X 2 E [(Bi+1 − Bu )(Bj+1 − Bv )] dudv+ E Bj+1 − Bj ds i,j=0 si sj j=0 sj In the first integral, we have, for j > i and u ∈ (i, i + 1) and v ∈ (j, j + 1): h i E Bi+1 Bj+1 − Bi+1 Bv − Bj+1 Bu + Bu Bv = (i + 1) − (i + 1) − u + u = 0 so the entire first integral is equal to 0, since the integrands are 0. (We used that E[Bs Bu ] = min(s, u). The second integral: !2 Z N −1 s j+1 X E Bj+1 − Bj ds = sj j=0 N −1 Z sj+1 X j=0 sj Z sj+1 sj h i E (Bj+1 − Bu ) (Bj+1 − Bv ) dudv Multiplying and calculating the integrand: h i 2 E Bj+1 − Bj+1 Bv − Bj+1 Bu + Bv Bu = sj+1 − v − u + min(u, v) ⇒ N −1 Z sj+1 X j=0 sj Z sj+1 sj (sj+1 − v − u + min(u, v)) dudv * Assuming now that s = u = v, we get back to the single integral: N −1 Z sj+1 N −1 Z sj+1 X X (sj+1 − 2s + s) ds = (sj+1 − s) ds j=0 sj j=0 Combining the summed integrals: Z 0 sj t (t − s)ds 5 ??????? ————————————————————————————————— Exercise 2 Consider the process dSt = σ(t)dBt , where σ(t) is a deterministic function being continuous and positive. This dynamics could be considered as the Bachelier model for stock prices, with time-dependent volatility. (a) Define martingale processes, and use this definition to show that St is a martingale. Formulate the martingale representation theorem, and show that St satisfies this. ————————————————————————————————— Definition 3.2.2 (Oksendal): Definition of Martingales. A stochastic process {Mt } is called a martingale wrt the filtration Ft if (i) Mt is Ft -measurable for all t. (ii) E[|Mt |] < ∞ for all t. (iii) s ≥ t, E[Ms |Ft ] = Mt . (Martingale property). Writing out St in the integral form. St = S0 + Z t σ(s)dBs . 0 (i) This does not take any values from after t, so it is Ft -adapted. Before verifying (ii), we need Cauchy-Schwarz’s inequality: E[|XY |] ≤ 1 1 E[X 2 ] 2 E[Y 2 ] 2 . Z t Z t E[|St |] = E S0 + σ(s)dBs ≤ E[|S0 |] + E σ(s)dBs 0 0 Applying Cauchy-Schwarz’s inequality on the second term with X = 1 and Y as the integral. "Z 2 # 21 Z t 21 t 2 ≤ E[|S0 |] + E σ(s)dBs = |S0 | + E σ(s) ds < ∞. 0 0 Since σ(s) is a continuous function it is bounded on the interval [0, t], so the integral is also bounded. Now it remains to show the martingale property, (iii). Assume t > s. Z t E[St |Fs ] = E S0 + σ(u)dBu Fs 0 6 S0 + E Z s 0 Z t σ(u)dBu Fs + E σ(u)dBu Fs s In the first expectation we have measurability, so we can factor the integral out of the conditional expectation. In the second term, we have a continuous, deterministic function with an elementary function representation. We define the partition: s = t0 < t1 < . . . < tN = t and write the integral as a sum: "N −1 # N −1 Z t i h X X E σ(u)dBu Fs = E ej E (Btj+1 − Btj ) Fs ej (Bj+1 − Bj ) Fs = s j=0 j=0 Since Brownian motion has independent increments, it is independent from the filtration and the conditional expectation becomes a normal expectation, and the normal expectation of ∆Bj is 0, so this terms vanishes. = S0 + Z s σ(u)dBu + 0 = S0 + 0 Z t σ(u)dBu = Ss . 0 Property (iii) verified: St is a martingale. Definition 4.3.4 (Oksendal): Martingale Representation Theorem. Suppose Mt is a martingale and Mt ∈ L2 (P ) (or E[Mt2 ] < ∞). Then there exists a unique, Ito integrable, stochastic process φ, such that Z t Mt = E[M0 ] + φ(s, ω)dBs 0 To apply the MRT to St we need to show that variance is finite. We have already shown it is a martingale. " 2 # Z t E[St2 ] = E S0 + σ(u)dBu = 0 Z t hZ t i2 2 E S0 + 2S0 σ(u)dBu + σ(u)dBu 0 0 Since the expectation is linear: Z t Z t 2 2 E S0 + 2S0 E σ(u)dBu + E σ(u)dBu 0 = S02 +0+ 0 Z t σ(u)2 du < ∞. 0 7 Hence, St satisfies the conditions of the MRT, so we can write: Z t St = E[S0 ] + φ(s, ω)dBs 0 or, S0 + Z t σ(s)dBs = S0 + 0 Z t φ(s, ω)dBs 0 so φ(s, ω) = σ(s). ————————————————————————————————— (b) Use Girsanov’s Theorem to find a probability Q such that exp(−rt)St is a Q-martingale. Here, r > 0 is the risk-free interest rate. You do not need to verify your change of probability. ————————————————————————————————— Using Girsanov’s Theorem, we get a new Brownian motion under Q: dWt = θt dt + dBt ⇐⇒ dBt = dWt − θt dt for some unknown θt . We get: dSt = σ(t) dWt − θt dt = −θt σ(t)dt + σ(t)dWt Lecture notes: It is a martingale measure if the discounted expected value is a martingale. We check with Ito: f (x, t) = xe−rt . d e−rt St = −re−rt St dt + e−rt dSt We use dSt under Q. = −re−rt St dt + e−rt − θt σ(t)dt + σ(t)dWt −rt =e − θt σ(t) − rSt dt + σ(t)e−rt dWt This is a martingale if the drift term is 0. −θt σ(t) − rSt = 0 =⇒ θt = − Using this for θt , we have the Q dynamics for St : dSt = − rSt dt + σ(t)dWt σ(t) 8 rSt σ(t) ————————————————————————————————— (c) Find St explicitly under Q, and show that St is normally distributed. Find mean and variance. ————————————————————————————————— Solving the SDE. rSt dt + σ(t)dWt σ(t) ————————————————————————————————— (d) Find the price at time zero, P0 of a digital option on St which pays 1 at time T if ST > K, and zero otherwise. Here, K > 0 is a constant. ————————————————————————————————— dSt = − ⌣ ¨ 9 ————————————————————————————————— Exercise 3 (a) State the Black & Scholes formula for call options. ————————————————————————————————— Theorem 4.3 (Benth): B&S Pricing Formula. For a call option (ST − K)+ the price Pt at time t is given by: Pt = St Φ(u1 ) − Ke−r(T −t) Φ(u2 ) where ln(St /K) + (r − 21 σ 2 )(T − t) √ u2 = σ T −t ln(St /K) + (r + 21 σ 2 )(T − t) √ , u1 = σ T −t where St is the stock price at time t, K is the strike price, T is the strike time, r is the interest rate, σ is the market volatility and Φ(·) is the normal cumulative distribution function. ————————————————————————————————— (b) What happens with the option price when the volatility tends to infinity or 0? ————————————————————————————————— As we can see from the formula above, the volatility only affects the parameters u1 and u2 . To simplify the notation we define a := ln(St /K) and τ := (T − t). First, when σ → ∞: a + rτ + 21 σ 2 τ a/σ + rτ /σ + 12 στ √ √ = lim = ∞. σ→∞ σ→∞ σ τ τ lim u1 = lim σ→∞ This means Φ(u1 ) → 1 as σ → ∞. a/σ + rτ /σ − 12 στ √ lim u2 = lim = −∞. σ→∞ σ→∞ τ Which means Φ(u2 ) → 0 as σ → ∞. From the formula of the option price we get: Pt = St (1) − Ke−r(T −t) (0) = St . The price of the option is always the same as the stock price. Now when σ → 0: rτ 1 √ a lim u1 = lim √ + lim √ + lim σ τ = ∞ σ→0 σ τ σ→0 2 σ→0 σ→0 σ τ | {z } | {z } | {z } →∞ →∞ 10 →0 Similarly, a rτ 1 √ lim u1 = lim √ + lim √ − lim σ τ = ∞ σ→0 σ→0 σ τ σ→0 σ τ σ→0 2 | {z } | {z } | {z } →∞ →∞ →0 so when σ → 0 both u1 → ∞ and u2 → ∞, so Φ(u1 ) = Φ(u2 ) = 1, and the option price becomes: Pt = St (1) − Ke−r(T −t) (1) = St − Ke−r(T −t) . Possibility of negative prices? No. Since the volatility is 0, we can perfectly predict the evolution of the stock price, so if the stock price would become less than the strike price with certainty, then the option would be worthless. 11