Advanced topics in continuous time finance

Transcription

1 Based on readings of Prof. Kerry E. Back on the IAS in Vienna, October 21. Advanced topics in continuous time finance Mag. Martin Vonwald November 21

2 Contents 1 Introduction Martingale Change of measure Security Price Model Proportional dividends Stock price density (Pricing kernel) Forward contracts Forward on a dividend paying security Stock options 1 4 International Finance 11 5 Definitions Variations Ito s Lemma Girsonov s Theorem Martingale representation theorem Valuation of stock options 18 7 Stochastic parameters Stochastic interest rates Stochastic volatility Risk neutral measure 25 2

3 Contents 9 Heston s model Heston s model under Q Futures contracts Foreign exchange Foreign risk neutral measure Quanto forward Straight forward on a foreign asset Wrong calculation (ignore this) Straight foreign forward Synthetic Forward Review stock options with proportional dividend yields Quanto option (no dividends) American put option Lower bound on put value Replicating value process Hedge of Quanto forward Examples Example Example Example Example Example Example

4 1 Introduction 1.1 Martingale (M t ) t is a martingale if: (M t ) t T is a martingale if: M s = E M t F s M t = E M T F t s t t 1.2 Change of measure Let x > a.s. Define Q (A) = 1 E x E x I A for all events A E Then Q is a probability measure equivalent to P meaning Q (A) = iff P (A) =. Write dq dp = X X Radon-Nikodyn derivative of Q with respect to P. Q (A) = E Q I A = 1.3 Security Price Model Z (t) =asset price A dq = 1 E (x) E x I x A = A E x dp D (t) =cumulative dividend process (assume: D (t) is increasing) Example: Proportional dividend dd (t) = δ (t) Z (t) dt D (t) = t δ (u) Z (u) du 4

5 1 Introduction S (t) =re-invested asset price Defintion: S (t) = Z (t) e t δ(u)du = Z (t) e t Write: θ (t) = e t 1 Z(u) dd(u) 1 Z(u) dd(u) 1.4 Proportional dividends dd (t) = δ (t) Z (t) dt θ (t) = e t δ(u)du θ (t) = θ (t) δ (t) dθ = θ (t) δ (t) dt S =value process of a self financing trading strategy. Example: money market account Suppose a risk free interest rate r Meaning there exists a security with price process S (t) = e rt Currency in a bank Z (t) = 1 S (t) = Z (t) e t rdu = e rt 1.5 Stock price density (Pricing kernel) Absence of arbitrage implies: there exists a stochastic process ρ such that for all T all reinvested asset processes S 1, S 2, t u and event A F t : { S2 (u) E ρ (T ) S 1 (T ) I A S 1 (u) S } 2 (t) = (1.1) S 1 (t) More fundamentally: for any self financing trading strategy θ with value process V θ it is valid: ρ =Arrow Debren state prices V θ () = E ρ (T ) V θ (T ) (1.2) 5

6 1 Introduction Proof of 1.1 from 1.2 We assume no dividends. Consider the following self financing strategy: In event A at time t buy one unit of security 2 and sell S 2(t) S 1 (t) units of security 1. At time u sell the unit of security 2 and buy S 2(u) S 1 (u) units of security 1. Hold until T : V θ (T ) = S 1 (T ) V θ () = This implies 1.1 (Arbitrage!). { S2 (u) S 1 (u) S } 2 (t) I A S 1 (t) Make the change of measure dq dp = ρ (T ) S 1 (T ) S 1 () E ρ (T ) S 1 (T ) = S 1 () { E I Q S2 (u) A S 1 (u) S } 2 (t) = S 1 (t) E Q S2 (u) S 1 (u) I A = E Q S2 (t) S 1 (t) I A S 2 (t) = E Q S2 (u) S 1 (t) S 1 (u) F t A F t, t u t u So it follows that is a Q-martingale. ( ) S2 (t) S 1 (t) t T Example S 1 (t) = e rt e rt S (t) = Q martingale S e rt S (t) = E Q e rt S (T ) Ft We say that we are using security 1 as the numeraire. 6

7 2 Forward contracts Lets define F (t) as the forward price at time t to receive X at time T. If you go long one forward contract at time t in event A F t you profit X F (t) I A : = E ρ (T ) X F (t) I A The goal is that is a Q-martingale. ( ) S2 (t) S 1 (t) t T Take S 1 to be a discount bond (zero bond) maturing at time T (S 1 (t) = e r(t t) ). If there exists a constant risk free rate then it follows: Now write: E ρ (T ) F (t) I A = E ρ (T ) X I A = S 1 () E S 1 () E ρ (T ) S1 (T ) S 1 () F (t) I A ρ (T ) S1 (T ) S 1 () X I A dq dp = ρ (T ) S1 (T ) S 1 () E Q F (t) I A = E Q X I A F (t) = E Q X F t We see that F is a Q-martingale. Changing the measure using a discount bond as numeraire is called the forward measure. If there is a constant risk free rate r then S 1(T ) S 1 () = ert. Using a discount bond as numeraire gives you some Q as using money market account. More generally if the risk free rate is deterministic then the price of a discount bond equals r(u)du S 1 (t) = e T t S 1 (T ) = e T r(u)du S 1 () 7

8 2 Forward contracts Example: Suppose you go long one forward at time and buy F () discount bonds. What is the value at time t? At time T you receive F (t) F (). The value at t is equal to F (t) F () S 1 (t) where S 1 (t) is the price of a discount bond maturing at time T. A portfolio including F () discount bonds is worth F (t) S 1 (t). Now check if this is valid at time T : V (T ) = X F (T ) S 1 (T ) = F (T ) }{{} = X 1 Making definitions more understandable X = E y F t Definition: E x I A = E y I A Write ε = y x so that y = x + ε Now take x to be F t measurable E y F t = x + E ε F t = x iff E ε F t = Definition: E y I A = E (x + ε) I A = E x I A + E ε I A = E x I A iff E ε I A = A F t We see that Define E ε F t = iff E ε I A = A F t E ε A = A εdp P (A) = E I A ε P (A) P (B A) = P (B A) P (A) = E I A I B P (A) = A I B (w) dp (w) P (A) E ε I A = A F t 8

9 2 Forward contracts We see that defining E ε F t = means that E ε A = A F t 2.1 Forward on a dividend paying security We assume that the dividends are deterministic and proportional. dd (t) = δ (t) Z (t) dt δ is non random S (t) = Z (t) e t δ(u)du The forward contract has a value of Z (T ) F () at time T. A Portfolio of a long forward and long F () discount bonds has a value of F () S 1 (t) or F () S 1 (T ) = Z (T ) at time T (S 1 is the price of the discount bond). An equivalent portfolio would be buying e T δ(u)du shares of the security at time. At time T we then have one share with value Z (T ). We see that if r is deterministic. e T δ(u)du Z () = F () S 1 () F () = 1 e T δ(u)du Z () S 1 () }{{} e T t r(u)du = e T (r(u) δ(u))du Z () 9

10 3 Stock options The value of an european call with strike price K and maturity T at time is equal to E ρ (T ) Z (T ) K + = E ρ (T ) Z (T ) I {Z(T )>K} + E ρ (T ) K I {Z(T )>K} }{{}}{{} : Use S 1 as price of a discount bond maturing at T E ρ (T ) K I {Z(T )>K} = S 1 () E ρ (T ) S1 (T ) S 1 () K I {Z(T )>K} = K S 1 () E Q I {Z(T )>K} = K S 1 () Q (Z (T ) > K) : Z (T ) equals the value of a portfolio of one forward on the security long and F () discount bonds long. The value of the portfolio at time is F () S 1 (): E Z (T ) ρ (T ) Z (T ) I {Z(T )>K} = F () S 1 () E ρ (T ) F () S 1 () I {Z(T )>K} = F () S 1 () E Q I{Z(T )>K} = F () S 1 () Q (Z (T ) > K) Specials case: deterministic proportional dividends F () S 1 () = e T δ(u)du Z () Z (T ) F () S 1 () = e T δ(u)du Z (T ) Z () = S (T ) S () Q is the measure obtained by using the reinvested asset price as numeraire. In particular, if δ =, then Q is obtained by using the stock price as numeraire. To really calculate the probabilities one has to be specific on the dynamics of S 1 and S 2 ; here comes the geometric brownian motion in. 1

11 4 International Finance Lets define X (t) as the exchange rate, i.e. the units of domestic currency per units of foreign currency. Let S f be a reinvested asset price in foreing currency. Then X (t) S f (t) equals the reinvested asset price in domestic currency. If S 1 is a domestic asset then X Sf S 1 is a Q-martingale, using S 1 as numeraire. { X (u) S f (u) E ρ (T ) S 1 (T ) I A X (t) } Sf (t) = A F t, t u T S 1 (u) S 1 (t) (4.1) Take S 1 (t) = X (t) S f 1 (t) for some foreign currency Sf 1. { } E ρ (T ) X (t) S f 1 (t) I S f (u) A S f 1 (u) Sf (t) S f 1 (t) = Now define ρ f as It follows: X () E ρ f (t) = ρ f (T ) S f 1 (t) I A ρ (t) X (t) X () { } S f (u) S f 1 (u) Sf (t) S f 1 (t) = (4.2) This looks like in chapter 2 on forwards. For all foreign reinvested assets prices S f 2 the expression Sf 2 (t) S f 1 (t) is a Qf -martingale where P is the same all over the world! dq f dp = ρf (T ) S f 1 (T ) S f 1 () Price of any foreign currency denominated claim y f at time T The value at time t = is equal to E ρ (T ) X (T ) y f (T ) ρ (T ) X (T ) = X () E y f (T ) X () 11

12 4 International Finance Now plug in formula 4.2 and we get X () E ρ f (T ) y f (T ) Foreign currency value at t = is E ρ f (T ) y f (T ) Example: Think of a foreign stock option with strike price set in foreign currency. No dividends are paid and the risk free rate is assumed to be constant. What is the value at t = in domestic currency? The simplest approach is to calculate (compare with chapter 3) X () E ρ f (T ) S f (T ) K f + = = X () F () S 1 () Q (S (T ) > K) K S }{{} 1 () Q (S (T ) > K) }{{} S() e rt Q is using the stock as numeraire; Q is using the money market account as numeraire. Example: Same as above, but now the strike price is in domestic currency. E ρ (T ) X (T ) S f (T ) K + = = E ρ (T ) X (T ) S f (T ) I {X(T ) S f (T )>K} E ρ (T ) K I {X(T ) S f (T )>K} The first part of the equation uses the foreign stock in foreign currency as numeraire; the second part uses the domestic discount bond as numeraire. 12

13 5 Definitions 5.1 Variations For a function f : R R the total variation of f is defined as i f The quadratic variation of f is defined as i f 2 i i If the total variation of f is finite, then the quadratic variation is equal to zero ( f has finite variation ). Suppose f is continously differentiable, then f (t) = f () + t and f has finite variations. We write df = f (x) dx. Suppose M is a continous martingale. Then either f (x) dx 1. with probability one the paths of M t M (t, ω) ω Ω have finite variations OR 2. M is constant a.s. Definition: We write the quadratic variation of M over, t as Then M 2 M, M is a martingale. M, M (t) 13

14 5 Definitions Definition: If M is a continous martingale then M is a Brownian Motion iff M, M (t) = t We write d M, M = dt or (dm) 2 = dt Definition: For two functions f, g : R R we define f, g = lim i i f i g If M and N are continous martingale then MN M, N is a martingale and M, N is called the covariance process of M and N. 5.2 Ito s Lemma Let dx = µ (t) dt + σ (t) dw with W being a Wiener process. If f = C 2 (i.e. it is a smooth, twice differentiable function) then it follows f (X t ) = f (X ) + X, X (t) = = t t t σ (u) 2 d W, W σ (u) 2 du f (X u ) dx u t d X, X (t) = σ (t) 2 dt df = f (x) {µdt + σdw} f (x) σ2 dt f (X u ) d X, X (u) Now let two processes X and Y be defined as If f is a smooth function it follows dx = µ X dt + σ X dw 1 dy = µ Y dt + σ Y dw 2 df (X (t), Y (t)) = δf δf dx + δx δy dy δ2 f δx 2 d X, X δ2 f δy 2 d Y, Y + δ2 f δxδy d X, Y d X, X = σxdt 2 d Y, Y = σ 2 Y dt d X, Y = σ X σ Y d W 1, W 2 d W 1, W 2 (t) = ρ (t) dt for some ρ < 1 14

15 5 Definitions 5.3 Girsonov s Theorem Let W be a Brownian motion on {Ω, F, P } and Q be a probability measure equivalent to P. Now define ξ = dq dp and set ξ (t) = E ξ F t. If we assume that dξ = λ (t) ξ (t) dw (t) for some λ then W (t) ˆ=W (t) + is a Brownian motion under Q. Further t λ (u) du dw = dw + λdt W, W = W, W = t 5.4 Martingale representation theorem Let the filtration (F t ) t be generated by the Brownian motion W. If M is a martingale, then there exists a Φ such that dm = ΦdW M (t) = M () + t Φ 1 (s) dw 1 (s) + We know that dξ = ΦdW and we write λ = Φ ξ. t Φ 2 (s) dw 2 (s) +... Equivalent statement The following statement is a Q-Brownian Motion: dw = dw 1 ξ d ξ, W We have so that d ξ, W = λ ξd W, W 1 d ξ, W = λdt ξ A Brownian motion W under P with up/down probability equal to 1 2 : 15

16 5 Definitions dt 2 dt Time dt 2 dt Now we look at W with drift: dt + λdt 2 dt + 2 λdt Time dt + λdt 2 λdt 2 dt + 2 λdt To make it a martingale the up probability must be 1 λ dt 2 = 1 λdw 2 and the down probability must be λ dt 2 = 1 λdw 2 First time step: Q 1 P 1 = 1 λdw 1 Q P = 1 λdw Second time step: Q 2 P 2 = (1 λdw 1 ) (1 λdw }{{} 2 ) Q 1 P 1 General: Q P Q P = Q 2 P 2 Q 1 P 1 Q 1 P 1 = λdw 1 Watch the sign! 16

17 5 Definitions From the theorem we see dξ ξ Q (A) P (A) = λdw = E ξ A 17

18 6 Valuation of stock options We assume a constant risk free rate r and no dividends. The value at time t = is equal to S Q (S (T ) > K) e rt K Q (S (T ) > K) }{{}}{{} where Q uses the stock as numeraire and Q uses the money market as numeraire. We further assume that ds S = µdt + σdw for constant µ and σ. : Under Q S e rt is a martingale (compare with chapter 5.2): d ( ) X Y d ( ) X Y X Y dx = µ X dt + σ X dw 1 dy = µ Y dt + σ Y dw 2 F (X, Y ) = X ( = S ) Y e rt δf δx = 1 Y δf δy = X Y 2 δ 2 F δ = 2 F = 2 X δx 2 δy 2 Y 3 = dx Y XdY Y 2 δf δxδy = 1 Y X Y 3 d Y, Y 1 Y 2 d X, Y = dx X dy Y + d Y, Y Y 2 d X, Y X Y d ( S e rt ) S e rt = ds S rdt = (y r) dt + σdw 1 Now set ξ = dq dp. By the martingale representation theorem dξ ξ = λdw for some λ By Girsonov s theorem we know that dw = dw + λdt 18

19 6 Valuation of stock options is a Q Brownian motion. d ( ) S e rt S = (µ r) dt + σ (dw λdt) e rt = (µ r σ λ) dt + σdw The difference of two martingales is again a martingale. So µ r σ λ = or if we rewrite the term This term is called the price of risk. We know that λ = µ r σ ds S = µdt + σdw = µdt + σ (dw + λdt) = rdt + σdw So we can calculate Q is unique! S (t) = S () e rt 1 2 σ2 t+σw (t) log S (t) = log S () + (r 12 ) σ2 t + σw (t) Q (S (T ) > K) = Q (log S (T ) > log K) ( = Q σw (T ) > log K log S () ( = Q W (T ) ) < d 2 T = N d 2 d 2 = log S () log K + ( r 1 2 σ2 T ) σ T (r 12 ) ) σ2 T : Now calculate S () Q (S (T ) > K) with Q using the stock as numeraire. 19

20 6 Valuation of stock options We know that e rt S(t) is a Q martingale and that for some λ the term dw = dw + λdt defines a Q Brownian motion. Further we know ds = µ Sdt + σ SdW d S, S = σ 2 S 2 dt Using Ito we calculate d ( ) e rt S(t) e rt S(t) = rdt ds S + d S, S S 2 = rdt µdt σdw + σ 2 dt = ( r µ + σ 2) dt σ (dw λdt) = ( r µ + σ 2 + σ λ ) dt σdw We conclude r µ + σ 2 + σ λ = ds = µdt + σ (dw λdt) S = ( µ + r µ + σ 2) dt + σdw = ( r + σ 2) dt + σdw log S (t) = log S () + (r + σ 2 12 ) σ2 t + σw (t) ( Q (S (T ) > K) = Q W ) (T ) < d 1 T = N d 1 d 1 = log S () log K + ( r σ2) T σ T 2

21 7 Stochastic parameters 7.1 Stochastic interest rates We know that ds S = µdt + σ (t) dw and that the money market account is equal to e t r(u)du. Discount bond maturing at T with bond price B and dynamics db = µ B dt + σ B dw. The forward measure uses the discount bond as numeraire. The second term in Black-Scholes: We calculate B () K Q (S (T ) > K) = E ρ (T ) K I {S(T )>K} e t r(u)du B (t) ( ) t e d r B t e r B = Q martingale = rdt db B + d B, B B 2 = ( r µ B + σ 2 B) dt σb dw Using dw = dw + λdt it follows Further we calculate r µ B + σ 2 B + σ B λ = ds = µdt + σdw S = µdt + σ (dw λdt) ( = µ σ µ B r σ 2 ) B dt + σdw σ B t ( log S (t) = log S () + µ σ µ B r σ 2 1 B σ B 2 σ2 ( T { = Q µ σ µ B + σ σ B 1 } σ B 2 σ2 du + T ) du + t σ σ B rdu + σdw T ) σdw > logk 21

22 7 Stochastic parameters The distribution of is given by In Vasicek N T T σ (t) dw (t) T, σ (t) 2 dt σ σ B rdt N 7.2 Stochastic volatility We are using Heston s model with a constant risk free rate and no dividends. ds S = µdt + vdw 1 v The distribution of v follows a mean-reverting CIR square root model whereas θ equals to the long run mean. dv = κ (θ v) dt + σ vdw 2 Value of a call option S () Q (S (T ) > K) e rt K Q (S (T ) > K) }{{}}{{} Q is using the stock as numeraire, Q is using the money market account as numeraire. : The term e rt S is a Q martingale d ( e rt S ) e rt = ds S S rdt = (µ r) dt + vdw 1 ξ = dq dp dξ ξ = λ 1 dw 1 λ 2 dw 2 dw 1 = dw 1 1 ξ d ξ, W 1 = dw 1 + λ 1 d W 1, W 1 + λ 2 d W 1, W 2 = dw 1 + λ 1 dt + λ 2 ρdt = (µ r) dt + v (dw 1 λ 1dt λ 2 ρdt) ρdt = d W 1, W 2 = const. 22

23 7 Stochastic parameters It must be µ r λ 1 v λ 2 ρ v = We have one equation but two unknowns (λ 1 and λ 2 ); we have an incomplete market and can t price via arbitrage. So we do equilibrium pricing, i.e. we are making some assumptions. ds S = µdt + vdw 1 = µdt + v (dw 1 λ 1dt λ 2 ρdt) = rdt + vdw 1 dv = κ (θ v) dt + σ vdw 2 = κ (θ v) dt + σ v (dw2 λ 1 ρdt λ 2 dt) dξ = λ 1 dw 1 λ 2 dw 2 ξ The state variable (volatility) is not traded! We are make assumption about λ 1 and λ 2 ; if we assume that λ 2 =, then λ 1 = µ r v. dv = {κ (θ v) dt (µ r) ρ σ} dt + σ vdw 2 = κ (θ v) dt + σ vdw 2 θ (µ r) ρ σ = θ θ > κ If we assume that λ 2 = we assume that the volatility risk is not priced. According to Hull&White If the path is known T Value of a call option: Q (S (T ) > K path of v) logs (T ) = logs () + r T 1 2 T vdw 1 N, vdt T T vdt + vdw 1 T T S () E N Q d 1 v = σ 2 T K e rt E N d 2 v = σ 2 T N d 1 and N d 2 are close to linear, so we approximate S () N d 1 K e rt N d 2 23

24 7 Stochastic parameters Where in d 1 put and in d 2 put if S () is close to K. E Q T E Q T vdt vdt for σ 2 T for σ 2 T 24

25 8 Risk neutral measure The money market measure is equal to the risk neutral measure Q. dq dp = ρ (T ) S1 (T ) S 1 () If S 1 is equal to the money market dq dp = ρ (T ) e T Let Q be the measure using the stock as numeraire: r(u)du ds S = µ S (t) dt + σ S (t) dw P whereas W P is a P Brownian motion. Now write S under Q: whereas W is a Q Brownian motion. ds S = rdt + σ S (t) dw dq = ρ (T ) S (T ) dp S () dq dq = dq dp dp dq = ρ (T ) S (T ) S () = e T r(u)du S (T ) S () 1 ρ (T ) e T r(u)du 25

26 8 Risk neutral measure Anything divided by the money market account S () is a martingale. E Q ( d e e e T rdu S (T ) S () ξ = dq dq ξ (t) ) = E Q ξ (T ) F t t r S t r S F t dξ ξ = rdt + ds S = σ S (t) dw = e t rdu S (t) S () = ξ (t) The term dξ ξ is always equal to the stochastic part of ds S. Recipe 1. Write model down under Q 2. Change to different numeraire 3. dw = dw With the stock as numeraire Write S under Q dξ ξ ( ) dξ ξ dw = σ S (t) dw = stochastic part of new numeraire dw = dw σ S dw dw = dw σ S dt ds = rdt + σ S dw S = rdt + σ S (dw + σ S dt) = ( r + σs 2 ) dt + σs dw W ˆ= Q Brownian Motion 26

27 9 Heston s model Under P ds = µdt + vdw 1 S dv = κ (θ v) dt + σ vdw 2 Now put ξ = dq dp and remember that W 1, W 2 = ρdt; by martingale representation: Under the risk neutral measure Q dξ ξ = λ 1 dw 1 λ 2 dw 2 dw1 = dw 1 dξ ξ dw 1 = dw 1 + λ 1 dt + λ 2 ρdt dw2 = dw 2 dξ ξ dw 2 = dw 2 + λ 1 ρdt + λ 2 dt ds S µ λ 1 v λ 2 ρ v = r = µdt + v (dw 1 λ 1 dt λ 2 ρdt) = ( µ λ 1 v λ 2 ρ v ) dt + vdw 1 Heston takes whereas λ is constant. Under Q he writes (λ 1 + λ 2 ρ) σ v = λ v dv = κ (θ v) dt + σ v (dw 2 λ 1 ρdt λ 2 dt) = κ (θ v) dt (λ 1 ρ λ 2 ) σ vdt + σ vdw 2 = κ (θ v) dt λ vdt + σ vdw 2 = κ (θ v) dt + σ vdw 2 κ = κ + λ θ = κ θ κ + λ 27

28 9.1 Heston s model under Q 9 Heston s model whereas W 1 and W 2 are Q Brownian motions. The value of an european call option is equal to ds = rdt + vdw 1 S dv = κ (θ v) dt + σ vdw 2 S () Q (S (T ) > K) e rt K Q (S (T ) > K) where Q uses the stock as numeraire. When we change from Q to Q Under Q We assume So we can write ξ = dq dq dξ = vdw 1 ξ dw1 = dw 1 dξ ξ dw 1 = dw 1 vdt dw2 = dw 2 dξ ξ dw 2 = dw 2 v ρdt ds = (r + v) dt + vdw1 S dv = κ (θ v) dt + σ ρ vdt + σ vdw 2 dw 1 dw 2 = ρdt d W 1, W 2 = ρdt S (t) = S () e rt 1 2 t vdu+ t vdw1 t log S (t) = log S () + rt 1 2 d log S (t) = (r 12 ) v dt + vdw 1 vdu + t vdw1 28

29 9 Heston s model Under Q d log S (t) = (r + 12 ) v dt + vdw1 ( ) κ θ dv = (κ σ ρ) κ σ ρ v dt σ vdw 2 f (x, v, t) = Q (S (T ) > K log S (t) = x, v (t) = v) = E Q I S(T )>K log S (t) = x, v (t) = v f (log S (t), v (t), t) ˆ= Q martingale As a martingale is a process such that we set We have defined f so that M (t) = E Q M (T ) F t M (t) = E M (T ) F t M (t) = f (log S (t), v (t), t) { 1 log S (T ) > log K M (T ) = otherwise df = δf δf δf d log S + dv + δx δv δt dt δ2 f δx 2 (d log S) δ2 f δv 2 (dv)2 + δ2 f (d log S) (dv) δxδv Drift of df under Q δf (r δx 12 ) v + δf δf κ (θ v) + δv δt δ2 f δx 2 v δ2 f δv 2 σ v + δ2 f δxδv σ v ρ = So we see that df is a martingale. We want: Q (S (T ) > K) = f (log S (), v (), ) It is important to check the boundary conditions: lim x f (x, v, t) = 1 lim t T f (x, v, t) = lim x f (x, v, t) = Further we look at Q (S (T ) > K) and set g (x, v, t) = E Q I S(T )>K log S (t) = x log v (t) = v { 1 x > log K x log K 29

30 9 Heston s model Drift of dg under Q Only the drift changes if we go from Q to Q! δg (r δx + 12 ) v + δg δv κ (θ v) + δg δt δ2 g δx 2 v δ2 g δv 2 σ v + δ2 g δxδv σ ρ v = We want the distribution of S (T ) under Q and Q. As a characteristic function identifies the distribution it suffices to find this characteristic funtion of log S: lnφ = E Q i Φ log S(T e ) Consider the following term: E Q e i d log S(T ) log S (t) = x log v (t) = v = k (x, v, t, Φ) For fixed Φ the term k (log S (t), v (t), t, Φ) is a Q martingale; it satisfies the PDE (df under Q). Now we guess a solution: log k (x, v, t, Φ) is an affine function of x and v with the parameters depending on t and Φ. 3

31 1 Futures contracts Define G (t) as the futures price for delivery of X at time T, whereas X is in currency units. If we buy a future at time and sell at time t we make a profit equal to G (t) G (). In fact the gain through t is Under Q: t e r (t s) dg (s) Therefore e rt t t d e r (t s) dg (s) = t e r s dg (s) e r s dg (s) = e r s dg (s) G (t) = E Q X F t F (t) = E Q X F t where Q is the forward measure using the discount bond as numeraire. If interest rates are deterministic Q is equal to Q, because T r dq dp = ρ (T ) e 1 dq 1 = ρ (T ) dp e T r = B () e T r 31

32 11 Foreign exchange Define X (t) as the exchange rate, i.e. the price of foreign currency in units of domestic currency. Use Q as the domestic risk neutral measure. For any foreign reinvested asset price S f the term e t rdu X (t) S f (t) is a Q martingale. Take S f for the money market account S f = e t rf du The term is a Q martingale. e t (r f r)du X (t) ( ) d e (r t f r)du X (t) e t (r f r)du X (t) = ( ) r f r dt + dx X So the drift of dx X = ( r r f) dt and we see dx ( X = r r f) dt + σ X (t) dw X (t) where W X is a Q Brownian motion. e t (r f r)du X (t) = E Q e T (r f r)du X (T ) F t X (t) = E Q e T t (r f r)du X (T ) F t Suppose that r f and r are constant. X (t) = E Q e (rf r) (T t) X (T ) F t = e (rf r) (T t) E Q X (T ) F t = e (rf r) (T t) G (t) where G (t) is a covered interest parity. 32

33 11 Foreign exchange 11.1 Foreign risk neutral measure We define Q as the foreign risk neutral measure, meaning that e t rf du S f (t) is a Q Brownian motion for all reinvested assets S. The domestic price is therefor given as E ρ (T ) X (T ) S f (T ) = X () E ρ (T ) X (T ) X () Sf (T ) = X () E ρ f (T ) S f (T ) ρ f (T ) = ρ (T ) X (T ) X () So 1 X() domestic price = foreign price = E ρ f (T ) S f (T ) and ρ f is a foreign state price density. The foreign risk neutral measure is given by Remember that Therefore Now set As the term can t have a drift, dξ ξ dq dp = ρf (T ) e T rf (t)dt dq dp = ρ (T ) e T r(t)dt dq dq = dq dp dp dq = ρ f e T rf dt 1 ρ (T ) e T = e T (r f r)dt X (T ) X () ξ = dq dq rdt ξ (t) = E Q ξ (T ) F t = E Q e T (r f r)dt X (T ) X () F t = e t (r f r)du X (t) X () is the stochastic part of dx X dξ ( ) ξ = r f r dt + dx X dξ ξ = σ X (t) dw X (t) 33

34 12 Quanto forward We are using the same pricing as in chapter 11.1 with the foreign asset S f and the forward price K in foreign currency. We fix the exchange rate at maturity as X, so the contract pays X (S f (T ) K ) in domestic currency at time T Straight forward on a foreign asset Pays X (S f (T ) K ) in domestic currency, where X is a contract-fixed exchange rate. Start under Q: dx ( X = r r f) dt + σ X dw X where W X is a Q Brownian motion. Know that is Q martingal. e rt X (t) S f (t) ds f S f = µ S dt + σ S dw S d ( e rt X (t) S f (t) ) e rt X (t) S f = rdt + dx (t) X + dsf S f + d X, S f }{{ X S f } where ρ is the correlation process of W X and W S. The drift of d(x Sf ) X S f must be rdt ( = rdt + r r f) dt + σ X dw X + µ S dt + σ S dw S + σ X σ S ρdt must be rdt. d ( X S f) X S f = dx X + dsf S f = + d X, S f X S f ( r r f) dt + σ X dw X + µ S dt + σ S dw S + σ X σ S ρdt Now compute the drift: ( ) r r f + µ S + σ X σ S ρ = r 34

35 12 Quanto forward It follows that µ S = r f σ X σ S ρ and so ds f S f = (r f σ X σ S ρ) dt + σ S dw S So we see that the drift of dsf S f is equal to the risk free rate minus the correlation of dx X with dsf S f. Now compute ) e rt E Q X (S f (T ) K = e rt X E Q S f (T ) e rt X K S f (T ) = S f () e (r f σ S σ X ρ 1 2 σ2 S) T +σ S W S (T ) E Q S f (T ) = S f () e (r f σ S σ X ρ) T Wrong calculation (ignore this) Pays S f (T ) K in foreign currency at time T. We start with a domestic point of view; the value of this contract is equal to e rt E Q X (T ) S f (T ) K = e rt E Q X (T ) S f (T ) e rt E Q X (T ) K = e rt X () S f () K e rt futures rate For this to be fairly priced K must be equal to K = spot rate futures rate Sf () Under the Q domestic risk neutral measure the value must be ) e rt E Q X (S f (T ) K = e rt X E Q S f (T ) e rt X K 35

36 12 Quanto forward Under the Q foreign risk neutral measure Further ds f S f = r f dt + σ S dw S dq dq = r) T e(rf X (T ) X () dq dq = e (r rf ) T X () X (T ) ξ = dq dq ξ = E Q ξ F t = e (r rf ) X () X (T ) dξ ( = r r f) dt dx d X, X + ξ X X 2 dws = dw S dξ ξ dw S = dw S σ X dw X dw S = dw S σ X γdt γ = d W X, W S ds f S f = r f dt + σ S (dws + σ X γdt) ( ) = r f + σ X σ S γ dt + σ S dws S f (T ) = S f () e (rf +σ X σ S γ 1 2 σ2 S) T +σ S WS (T ) E Q S f (T ) = S f () e (rf +σ X σ S γ) T THIS IS WRONG! WRONG SIGN! AD PROBLEMS: K Z (T ) + = 1 K Z (T ) Z (T ) 1 K E t X T = E X t F t Straight foreign forward Forward contract on a foreign asset (no dividends) with price in foreign currency (K =forward price). 36

37 12 Quanto forward Under the domestic risk neutral measure, where X S f is a domestic asset, we want E Q X (T ) S f (T ) K = Compute E Q X (T ) S f (T ) Because is a Q martingale it follows e rt X (t) S f (t) X () S f () = E Q e rt X (T ) S f (T ) e rt X () S f () = E Q X (T ) S f (T ) Compute E Q X (T ) K E Q X (T ) K = K E Q X (T ) = K G () where G () is the currency futures price at time. Covered interest parity: X () = e (rf r) T G () E Q X (T ) K = K e (r rf ) T X () Now we know e rt X () S f () K e (r rf ) T X () = K = e rt S f () 12.3 Synthetic Forward Construct a portfolio, where we buy the asset and borrow S f () in foreign currency: we owe e rt S f () at time T. Start under the domestic risk neutral measure Q. 1. Stock as numeraire Q ds S = rdt + σ S dw s ξ = dq dq under Q dξ ξ = σ SdW S 37

38 12 Quanto forward For any other Q Brownian motion W in the model we define a Q Brownian motion as dw = ( ) ds dw (dw ) ξ = dw σ S (dw s ) (dw ) = dw σ S γdt for some γ, which describes the relation to the stock. 2. Forward measure Q (uses discount bond as numeraire) db B = rdt + σ BdW B ξ = dq dq under Q dξ ξ = σ BdW B For any other Q Brownian motion W we define a Q Brownian motion as dw = ( ) ds dw dw S = dw σ B γdt for some γ. 3. Foreign risk neutral measure Q e rf t S f (t) is a Q martingal for any foreign reinvested asset price S f. Under Q dx X = ( r r f) dt σ X dw x ξ = dq dq dξ ξ = σ XdW X For any other Q Brownian motion W we define a Q Brownian motion as for some γ. Given two Brownian motions, W 1 and W 2, dw = dw σ X γdt Then the term d (W 1 ) (W 2 ) = d W 1, W 2 n W 1, W 2 (t) = lim partitions ( W 1 ) ( W 2 ) W 1 (t) W 2 (t) W 1, W 2 (t) 1 38

39 12 Quanto forward is a martingale. Suppose: cov (W 1 (t), W 2 (t)) = E W 1 (t) W 2 (t) = E W 1 (t) E W }{{} 2 (t) }{{} For some ρ This means d (W 1 ) d (W 2 ) = ρdt d W 1, W 2 (t) = ρdt W 1, W 2 (t) = W 1, W 2 () + }{{} t ρ (s) ds So the term is a martingale. E W 1 (t) W 2 (t) W 1 (t) W 2 (t) t t ρ (s) ds ρ (s) ds = t cov (W 1 (t), W 2 (t)) = E ρ (s) ds As the standard deviation of W (t) equals t Now consider corr (W 1 (t), W 2 (t)) = cov (W 1 (t), W 2 (t)) σ 1 σ 2 1 t = E ρ (s) ds t dm 1 = σ 1 dw 1 dm 2 = σ 2 dw 2 This means M 1 (t) = M 1 () + M 2 (t) = M 2 () + t t σ 1 (s) dw 1 (s) σ 2 (s) dw 2 (s) 39

40 12 Quanto forward Write (dm 1 ) (dm 2 ) σ 1 σ 2 ρdt where ρ is the correlation process for W 1 and W 2. This means and so is a martingale. d M 1, M 2 = σ 1 σ 2 ρdt M 1, M 2 (t) = t M 1 M 2 M 1, M 2 t cov (M 1 (t), M 2 (t)) = E t var (M 1 (t)) = E σ 1 (s) σ 2 (s) ρ (s) ds σ 1 (s) σ 2 (s) ρ (s) ds σ1 2 (s) ds 4

41 13 Review stock options with proportional dividend yields Let Z be the asset price and δ a constant proportional dividend yield: Now compute dd = δ Z S (t) = e δt Z (t) e rt E Q (Z (T ) K) t = e rt E Q Z (T ) I Z(T )>K e rt K E Q I Z(T )>K ds S dz Z E Q I Z(T )>K = rdt + σdw = (r δ) dt + σdw = Q (Z (T ) > K) = N d 2 S() log K d 2 = + ( r δ 1 2 σ2) σ T E Q Z (T ) I Z(T )>K = e δt E Q S (T ) I Z(T )>K e = e (r δ)t S () E Q rt S (T ) I S () Z(T )>K dq = e (r δ)t S () E Q dq I Z(T )>K = e (r δ)t S () E Q IZ(T )>K = e (r δ)t S () Q (Z (T ) > K) Switching from Q to Q adds σ 2 dt to the drift of dz Z : e rt E Q Z (T ) I Z(T )>K = e δt S () N d 1 So the formula for a call equals to: d 1 = log S() K e δt S () N d 1 e rt K N d 2 + ( r δ σ2) T σ T 41

42 13 Review stock options with proportional dividend yields 13.1 Quanto option (no dividends) Pays X (S f (T ) K ) +. The option value is equal to e rt E Q X ( ) + S f (T ) K ds f S f = ( = e rt X ) + E Q S f (T ) K ( ) r f σ X σ S ρ dt + σ S dw S = (r δ) dt + σ S dw S δ = r + σ X σ S ρ r f So we can compute the value of the quanto option as: X e δt S () N d 1 X e rt K N d American put option The optimal exercise time is the first time that S (t) drops to the optimal exercise boundary b (t), which is a monotonic increasing function with lim t T K, i.e the exercise accelerate the receipt of K. The value of an american put option is equal to sup E Q e rτ (K S (τ)) + = E e rτ (K S (τ )) + τ where τ is the exercise time and τ (ω) = t means exercise at t in state ω of the world and τ = inf {t S (t) b (t)}. We can think of the value of an american put options as the value of an european put option plus some kind of early exercise premium: E Q T e ru I S(u) b(u) r Kdu Now set P (x, t) equal the put value at time t where S (t) = x. We know that P (x, t) = K x when x b (t). If x > b (t) then P (x, t) is equal to the european value at t when S (t) = x + exercise premium: Our value matching condition is T E Q e r(u t) I S(u) b(u) r Kdu S (t) = x t K b (t) = {european put value at t when S (t) = b (t)} + T +E Q e r(u t) I S(u) b(u) r Kdu S (t) = b (t) t 42

43 13 Review stock options with proportional dividend yields which gives us the integral condition to be solved for the function b (see Broadie and Detemple, Review of financial studies, 1996 for a fast numerical procedure). The early exercise premium at t is equal to T t r K e r(u t) Q (S (u) b (u) S (t)) = Now consider different constant exercise boundaries: T r K e r(u t) N t d u t, S (t), b (u) }{{} role of K ds = rdt + σdw S S (u) = S (t) e (r 1 2 σ2 ) (u t)+σω(u) ω(t) log S (u) = log S (t) + (r 12 ) σ2 (u t) + σ ω (u) ω (t) ( Q (log S (u) < log b (u)) = Q log S (t) + (r 12 ) ) σ2 (u t) + σ (ω (u) ω (t)) < log b (u) ω (u) ω (t) = Q u t b(u) log S(t) ( r 1 2 σ2) (u t) σ u t }{{} d(u t,s(t),b(u)) How to find the lower boundary Dont work with optimal boundaries but assume constant (i.e. flat lines) boundaries. 1. Find the best flat line (simpler than curved lines). 2. Hypothetically move this line down and start again. 3. End up closely above intersecting b. 4. Find this point for every t. That is the lower boundary. Now find boundary c (t) that is slightly above b (t): T t r K e r(u t) N d (u t), S (t), c (u) > true early exercise premium 43

44 13 Review stock options with proportional dividend yields Lower bound on put value Consider a flat boundary X starting at t. Now calculate E e r(t t) (K S (τ)) + S (t) where τ = inf {t S (t) X} The hitting time of a flat line by a brownian motion has a known distribution. If we maximize over X we get the lower bound. 44

45 14 Replicating value process Let the assets Y and Z be reinvested assets. Under Q we have dy Y = rdt + σ Y dw Y dz Z = rdt + σ ZdW Z Let V be the value of a self financing contingent claim, e.g. V (t) = e r (T t) E Q X F t Suppose that V can be replicated by using Y and Z, i.e. V (t) = f (t, Y (t), Z (t)) for some f Generally we have According to Ito s Lemma: dv V = rdt + stochastic part 1. Derive fundamental PDE (Drift of df = r fdt ) 2. The stochastic part of dv is equal to δf δf δz stochastic part of dy + δz stochastic part of dz. It follows: df = r f dt + δf δf (dy r Y dt) + (dz r Zdt) δy δz = δf ( δf dy + δy δz dz + f δf δy Y δf ) δz Z rdt Now build a portfolio with δf δf δf δy shares of Y and δz shares of Z and f market, which will replicate f, i.e. change in value of portfolio equals df. δy δf δz in the money 45

46 14 Replicating value process 14.1 Hedge of Quanto forward The contract pays X (S f (T ) K ) at time T. The fair price K is such that the contract has value at time : ) e rt E Q X (S f (T ) K = E Q S f (T ) = K Now consider X S f : dx X + dsf S f The fair value at time t must be equal to ds f S f = ( ) r f σ X σ S ρ dt + σ S dw S d ( X S f ) X S f = rdt + stochastic part + d X, S f X S f = rdt + stochastic part = (r r f) dt + dsf S f + σ X σ S ρdt } r = r r f + σ X σ S ρ + {dt part of dsf S f S f (T ) = S f () e (rf σ X σ S ρ 1 2 σ2 S) T +σ S W S (T ) E Q S f (T ) = S f () e (rf σ X σ S ρ) T = fair value at time F (t) = S f (t) e (rf σ X σ S ρ) (T t) Now consider a long forward at time at price F (). The value at time t is equal to: e r(t t) E Q X (F (t) F ()) = e r(t t) If you sell a forward at time and you want to hedge, then you need a portfolio with value e r(t t) X (F (t) F ()) = e r(t t) X ( ) e (rf σ X σ S ρ) (T t) S f (t) F () As S f is non-domestic it is tricky to calculate. We set = e ( r+rf σ X σ S ρ) (T t) X S f (t) e r(t t) X F () Y (t) = X (t) S f (t) Z (t) = e rf t X (t) 46

47 14 Replicating value process It follows Y Z = e rf t S f (t) S f (t) = e rf t Y (t) Z (t) Now set g (t, Y, Z) = e ( r+rf σ X σ S ρ) (T t) X e rf T Y Z e r(t t) X F () = e ( r+rf σ X σ S ρ) (T t) X S f (t) e r(t t) X F () Then g is the value process we want to replicate. δg δy δg δz = number of shares of Y to hold = number of shares of Z to hold Example: δg δy = Z δg δy g δg δy Y δg δz Y = value of Y shares in domestic currency Z = invested in domestic money market δg δy = e ( r+rf σ X σ S ρ) (T t) X e rf t = X X (t) e( r+rf +σ X σ S ρ) (T t) δg δy Y (t) = e( r+rf σ X σ S ρ) (T t) X S f (t) δg δy δg δy X (t) Sf (t) = δg = number of shares of foreign asset to hold 1 e rf t (X (t)) δy Y (t) value of foreign asset in domestic currency δg δg Z (t) = δz δz t erf X (t) = Value in domestic currency of position in foreign money market 47

48 15 Examples 15.1 Example 1 Consider e european put option on a stock with constant proportional dividend yield δ and constant volatility coefficient σ. Assume the risk-free rate is constant. The payoff of the put, with exercise price K, is K Z (T ) + 1 = K Z (T ) Z (T ) 1 + K The interpretation is that exchanging one share of K units of currency is equivalent to buying the currency at a price equal to one share, or equivalent to K purchases of a currency unit, each 1 at cost K when measured in shares. The price of a currency unit in shares is of course 1 Z(T ). Use this formula and a change of measure to show that the value of the put is given by the Black- Scholes call option formulae when we interchange Z () with K and interchange r with δ in the formula. Note that this is not the usual put-call parity! Solution: At time we have 1 E γ (T ) K Z (T ) Z (T ) 1 + K 1 = E γ (T ) K Z (T ) Z (T ) I 1 Z(T ) > 1 K }{{} E S (T ) K Z (T ) 1 K I 1 Z(T ) > 1 K } {{ } Using the discount bound as numeraire (S 1 ˆ=money market) we get for : S 1 () E γ (T ) S1 (T ) S 1 () K I Z(T )<K = K S 1 () Q (Z (T ) < K) And with F for the forward price of the stock and with a long forward on the stock together with F () discount bonds as numeraire we get for : F () S 1 () Q (Z (T ) < K) 48

49 15 Examples Q (Z (T ) < K) = 1 Q (Z (T ) K) = N d 2 log K S() = N + ( δ r σ2) T σ T Q (Z (T ) < K) = N d 1 log K S() = N + ( δ r 1 2 σ2) T σ T 15.2 Example 2 Suppose the domestic and foreign interest rates are constant and equal. 1. Show that under the domestic risk-neutral measure Q, E Q t X (T ) F t = X (t), where X is the exchange rate (price of foreign currency in units of domestic currency). Start from the most primitive assumptions you can. Solution: S f (t) = e S f (t) X (t) e t rdu ˆ= Q martingal t rf du X (t) e t (r f r)du = X (t) 2. Show that unless X is constant. Solution: E Q t 1 X (T ) 1 X (t) Under Q is must be dx X = dt + σ X (t) dw Now we use Ito with f (t, X) = 1 X df = δf δt = δf δx = 1 X 2 δ 2 f δx 2 = 2 1 X 3 ( ) 1 2 σ2 X (t) 1 X2 2 X 3 dt + σ x (t) X 1 X 2 dw = σ 2 X 1 X dt σ X (t) 1 X dw d 1 X 1 X = σ 2 X (t) dt σ X (t) dw 49

50 15 Examples So we see that the drift of d 1 X 1 X is zero iff X is constant, i.e. σx 2 (t) is zero Example Example 4 A supershare pays S(T ) K, when K 1 S (T ) K 2 and nothing otherwise, where K 1 and K 2 are given. What is the fair value of a supershare if the interest rate is constant and S is a nondividend paying asset with constant volatility? What if S pays a constant proportional dividend yield? Solution: 1. Without dividend S (T ) V = e rt E Q I K K1 S(T ) K 2 1 = e rt K 1 E Q S (T ) I K1 S(T ) K 2 Now change measure and use the stock price as numeraire: Under Q we have V = S () K 1 E Q e rt S (T ) S () } {{ } dq dq = S () K 1 E Q IK1 S(T ) K 2 I K1 S(T ) K 2 = S () K 1 {Q (S (T ) K 1 ) Q (S (T ) K 2 )} ds S = rdt + σdw Q dw Q = dw Q dξ ξ dw Q = dw Q σ dw Q dw Q }{{} dt 5

51 15 Examples Under Q we have then ds S = ( r + σ 2) dt + σdw Q S (t) = S () e (r+σ2 1 2 σ2 )t+σw Q log S (t) = log S () + (r + 12 ) σ2 t + σw Q W Q N, t So the fair value of the share is equal to V = S () ( (Q log S () + K 1 = S () K 1... Q = S () K 1 (N dk 2 N dk 1 ) (r + 12 σ2 ) ) ) T + σw Q (T ) log K 2... W Q (T ) log K 1 S() ( r σ2) T T σ T 2. With dividend we only have to substitute r with r δ. } {{ } d(k 1 ) 15.5 Example 5 Show that under the same assumptions as in problem 2 that 1 E Q X (T ) F t = 1 X (t) where Q is the foreign risk-neutral measure. Solution: Using S (t) = e t rdu under Q we get S (t) 1 X (t) e t rf du ˆ= Q martingale = e t (r r f )du 1 X (t) If r = r f this equals to 1 X(t). 51

52 15 Examples 15.6 Example 6 Consider the stochastic volatility model ds S = rdt + vdw 1 dv = κ (θ v) dt + σdw 2 written under Q, where d W 1, W 2 =. Derive PDE s for the two probabilities in the call option pricing formula. Solution: 1. Find d 1, so that Q (P (T, u) > K) = N d 1 where Q uses the u-maturity bond as numeraire. We know that When we set dq dq σ (t, s) = σ ( dp (t, u) u ) = rdt σ (t, s) ds dw (t) P (t, u) t = rdt (u t) σdw (t) = ξ, we have dξ ξ = (u t) σdw (t) and dw = ( ) dξ dw dw ξ = dw + (u t) σdt defines a Q -Brownian motion. Substituting gives us dp (t, u) P (t, u) = rdt (u t) σdw (t) + (u t) 2 σ 2 dt Also implies dw = dw + (u t) σdt W (t) = W (t) + = W (t) + t (u s) σds ( ) u t t2 σ 2 52

53 15 Examples So we get ( ) r (t) = r () + θ (t) + σ W (t) u t t2 σ 2 2 ( ) Now we set φ (t) = θ (t) u t t2 2 σ 2 and rewrite r (t) as r (t) = r () + φ (t) + σw (t) The formula for dp P implies P (T, u) = P (, u) e T r(t)dt+σ2 = P (, u) e T r(t)+ σ2 2 T (u t)2 dt σ T (u t)2 dt σ T (u t)dw (t) 1 2 σ2 T (u t)dw (t) = P (, u) e T r()+ T φ(t)dt+σ T W (t)dt+ σ2 2 Apply Ito s Lemma to the function f (t, W ) = t W and we get Hence, σ T T W (T ) W () = T (u t)2 dt σ df = W (t) dt + t dw (t) = = T T T df W (t) dt tdw (t) T (u t) dw (t) = σ u W (T ) σ = σ u W (T ) + σ T = σ (u T ) W (T ) + σ Substituting in the formula for P (T, u) gives P (t, u) = P (, u) e T r()+ T φ(t)dt+ σ 2 2 tdw (t) T (u t)2 dt T (u t)dw (t) W (t) dt σ T W (T ) T W (t) dt T u t+ t2 2 + u2 2 t u+ t2 2 dt Also T φ (t) dt + σ2 2 T (u t) 2 dt = = = T T T T θ (t) dt + σ 2 t22 u2 ( u t t u + t2 2 θ (t) dt + σ 2 ( u T 2 + T u2 T T 2 ) u 2 2 ( θ (t) dt σ 2 u T 2 u2 T T 3 ) 2 3 ) dt 53

54 15 Examples We have P (T, u) > K iff log P (T, u) > log K: T ( log P (, u) + T r () + θ (t) dt σ 2 u T 2 u2 T 2 log P (, u) K + T r () + T θ (t) dt σ 2 T 3 ) σ (u T ) W (T ) > log K 3 ( u T 2 u2 T 2 T 3 ) 3 To convert σ (u T ) W (T ) to a standard normal, divide by σ (u T ) T. Hence { 1 d 1 = σ (u T ) T P (, u) T ( log K + T r () + θ (t) dt σ 2 u T 2 u2 T T 3 )} Find d 2, so that Q P (T, u) > K = N d 2 where Q uses the T -maturity bond as numeraire. The dynamics of the T -maturity bond are so dp (t, T ) P (t, T ) defines a Q -Brownian motion. This gives = rdt (T t) σdw (t) dw = dw + (T t) σdt t W (t) = W (t) + (T s) σds ( ) = W (t) + σ T t t2 2 > σ (u So r (t) = r () + θ (t) σ 2 ( ) T t t2 + σ W (t) 2 The option is written on the u-maturity bond and dp (t, u) P (t, u) = rdt (u t) σdw (t) = rdt + σ 2 (T t) (u t) dt (u t) σdw (t) So P (T, u) = P (, u) e T r(t)dt+σ2 = P (, u) e T r()+ T θ(t)dt σ2 T (T t) (u t)dt σ T (u t)dw (t) σ2 2 T (u t)2 dt 2 T 3 3 u2 T 2 σ (u T ) W (T ) Thus P (T, u) > K iff T ( log P (, u)+t r ()+ θ (t) dt σ 2 2 T 3 3 u2 T 2 ) σ (u T ) W (T ) > log K 54

55 15 Examples So we define d 2 as { 1 d 2 = σ (u T ) T P (, u) T ( )} 2 T log K + T r () + θ (t) dt σ 2 3 u2 T The value of the options is therefore E Q e T r(t)dt P (T, u) I P (T,u)>K E Q e T r(t)dt K I P (T,u)>K = = P (, u) Q (P (T, u) > K) K P (, T ) Q (P (T, u) > K) = P (, u) N d 1 K P (, T ) N d 2 55