Continuous-Time Finance 1 / 46 Self-financing portfolio of stocks and bonds Discrete time: At period t, investor holds portfolio of value Vt , consisting of... ∆t shares of the stock, priced at St per unit, and Vt − ∆t St units of account invested in one-period bonds. Portfolio value at t + 1 is Vt+1 = ∆t St+1 + Rf,t (Vt − ∆t St ) . Subtract Vt from both sides to get Vt+1 − Vt = ∆t (St+1 − St ) + (Rf,t − 1) (Vt − ∆t St ) . | {z } | {z } | {z } ≈dVt ≈dSt ≈rt dt Continuous time analogue: dVt = ∆t dSt + rt (Vt − ∆t St )dt. 2 / 46 Admissible portfolio Definition A self-financing portfolio is admissible if the corresponding wealth process {Vt } is bounded below, i.e., ∃v̄ < ∞ s.t. Vt ≥ −v̄, ∀ (t, ω) ∈ [0, T ] × Ω. Intuitively, there must be limit to how much debt can be tolerated. Without admissibility, can generate any VT > 0 (with probability 1) from V0 = 0. Recall that a borrowing limit is required to rule out the discrete-time doubling strategy. 3 / 46 Replication Pricing & Black-Scholes 4 / 46 Replication pricing in continuous time Idea is same as in discrete time: Find self-financing portfolio that perfectly replicates derivative payoff. Demonstrate idea by pricing options in the Black-Scholes setting: Return on money market account is constant r, i.e., its value β is βt = ert . Stock price, S, follows a geometric BM, i.e., dSt = µSt dt + σSt dBt . 5 / 46 Option value Consider European call option that pays (ST − K)+ at time T . Conjecture that option price at t should depend on: Time to expiration, T − t, and Current price of underlying stock, St . (Also on parameters, e.g., strike K, but these don’t vary over time) Why is this reasonable? Stock price dynamics (dSt ) only depend on itself (St ). Option payoff depends only on expiration stock price (ST ). Time matters because of potential for shocks to hit the stock price before expiration; more things can happen with more time left. Thus, we write the call value, Ct , as function of t and St : Ct = c(t, St ). Note that the function c itself isn’t random, but Ct is, since St is random. 6 / 46 Replicating portfolio Assets: start with wealth V0 ; make self-financing trades in stock & risk-free asset. Goal: want to make sure Vt = Ct for all times 0 < t ≤ T . Match the terminal value: VT = CT . Match all the dynamics: dVt = dCt for all t < T . Result: if so, then we can price the call by V0 = C0 . Z T Z T CT − C0 = dCt = dVt 0 0 = VT − V0 = CT − V0 ⇒ V 0 = C0 . (by the same logic Vt = Ct for all times t) Equivalent method: Match d(Vt /βt ) = d(Ct /βt ) for all t. Determines the portfolio of stocks and bonds that we need to hold. Need to derive SDEs for Vt /βt and Ct /βt . 7 / 46 SDE for discounted portfolio value 1. Write Vt /βt = e−rt Vt as f (t, Vt ) for f (t, v) = e−rt v. 2. To use Itô’s lemma, note ft = −re−rt v, fv = e−rt , and fvv = 0, so 1 d(Vt /βt ) = ft dt + fv dVt + fvv Vart [dVt ] 2 −rt −rt = −re Vt dt + e dVt = e−rt (dVt − rVt dt). 3. Since Vt is self-financing, dVt = ∆t dSt + r(Vt − ∆t St )dt, so e−rt (dVt − rVt dt) = e−rt ∆t (dSt − rSt dt). 4. Finally, use formula for geometric BM, so h i d (Vt /βt ) = e−rt ∆t (µ − r) St dt + σSt dBt . 8 / 46 SDE for discounted option price 1. Use Itô’s lemma on Ct = c(t, St ) and plug in formula for geometric BM to get 1 dCt = ct dt + cS dSt + cSS Vart [dSt ] 2 1 2 2 = ct + µSt cS + σ St cSS dt + σSt cS dBt . 2 2. For Ct /βt , use Itô’s lemma (exactly as in previous slide) to find immediately: d(Ct /βt ) = e−rt (dCt − rCt dt). 3. Then substitute from the above (and substitute Ct = c(t, St )) to get: h i 1 d(Ct /βt ) = e−rt (−rc + ct + µSt cS + σ 2 St2 cSS )dt + σSt cS dBt . 2 9 / 46 Deriving the Black-Scholes PDE Using the previous two slides, the condition d(Vt /βt ) = d(Ct /βt ) becomes h i 1 ∆t (µ−r)St dt+σSt dBt = −rc+ct +µSt cS + σ 2 St2 cSS dt+σSt cS dBt . 2 Equating the dBt terms, we solve for the option delta: ∆t = cS , ∀t ∈ [0, T ) . Remember the formula from the binomial tree, ∆t = u −C d Ct+1 t+1 . u −S d St+1 t+1 Equating the dt terms, and plugging ∆t , we have 1 rc = ct + rSt cS + σ 2 St2 cSS , 2 ∀t ∈ [0, T ). In conclusion we seek function c(t, S) satisfying partial differential equation 1 rc(t, S) = ct (t, S) + rScS (t, S) + σ 2 S 2 cSS (t, S), ∀t ∈ [0, T ), S ≥ 0, 2 with the terminal condition c(T, S) = (S − K)+ . 10 / 46 Aside: alternative derivation of Black-Scholes PDE 1. Apply Itô’s lemma on Ct = c(t, St ) as before to get 1 dCt = ct dt + cS dSt + σ 2 St2 cSS dt. 2 2. Rearrange to get 1 = ct + σ 2 St2 cSS dt . dCt − cS dSt {z } | 2 {z | } capital gains of portfolio long 1 call, short cS stocks risk-free return since no dBt term 3. By no-arbitrage, equate (ct + 12 σ 2 St2 cSS )dt to capital gains of investing Ct − cS St in risk-free asset, i.e., 1 r c − cS St dt = ct + σ 2 St2 cSS dt. 2 4. Divide by dt and rearrange to get the same PDE as the previous slide: 1 rc = ct + rSt cS + σ 2 St2 cSS . 2 11 / 46 Black-Scholes formula How do we solve the PDE to get an expression for c(t, S)? Typically, we would proceed numerically, by using standard techniques for solving PDEs. In this case, there is a nice closed-form solution: (Φ is the standard normal cdf) log(S /K) + (r + 1 σ 2 )(T − t) t 2 √ St Ct = Φ σ T −t log(S /K) + (r − 1 σ 2 )(T − t) t 2 √ −Φ Ke−r(T −t) σ T −t It’s a huge pain, but you could check that this satisfies the PDE by taking derivatives wrt t, S, and SS, and plugging them into the PDE 12 / 46 “The Greeks” Ct = Φ log(St /K) + (r + 1 σ 2 )(T − t) log(St /K) + (r − 1 σ 2 )(T − t) 2 2 √ √ St − Φ Ke−r(T −t) σ T −t σ T −t | {z } | {z } :=x1 :=x2 e.g., “Delta” ∆t := h i ∂Ct 1 √ = Φ(x1 ) + φ(x1 )St − φ(x2 )Ke−r(T −t) ∂St {z } σSt T − t | =0 (just takes a bit of algebra) = Φ(x1 ) Some others are φ(x1 ) ∂ 2 Ct √ = ∂St2 σSt T − t √ ∂Ct = φ(x1 )St T − t “Vega” Vt := ∂σ h i ∂Ct ∂Ct σ “Theta” Θt := =− = Ke−r(T −t) rΦ(x2 ) + √ φ(x2 ) ∂(T − t) ∂t 2 T −t “Gamma” Γt := 13 / 46 Call price Ct is a convex function of St (because Γt > 0); also Ct > CT (why?) 0.3 0.25 0.2 0.15 0.1 0.05 0 0.75 0.8 0.85 0.9 0.95 1 1.05 1.1 1.15 1.2 1.25 14 / 46 Call price Ct increases with volatility σ (because Vt > 0) 0.3 0.25 0.2 0.15 0.1 0.05 0 0.75 0.8 0.85 0.9 0.95 1 1.05 1.1 1.15 1.2 1.25 15 / 46 Call price Ct increases with time-to-maturity T − t (because Θt > 0) 0.3 0.25 0.2 0.15 0.1 0.05 0 0.75 0.8 0.85 0.9 0.95 1 1.05 1.1 1.15 1.2 1.25 16 / 46 Put options Can recover Black-Scholes for a put option by simple put-call parity: Ct − Pt + Ke−r(T −t) = St Thus, Pt = Φ(x1 )St − Φ(x2 )Ke−r(T −t) +Ke−r(T −t) − St {z } | =Ct = [1 − Φ(x2 )]Ke−r(T −t) − [1 − Φ(x1 )]St = Φ(−x2 )Ke−r(T −t) − Φ(−x1 )St 17 / 46 Risk-Neutral Pricing 18 / 46 Motivation Goal is to price a financial instrument (e.g., option). We saw above the PDE approach. But it is sometimes complicated. From discrete-time, know can write price as risk-neutral expectation h i + Ct = e−r(T −t) EQ (S − K) . T t But to calculate Q-expectation, need to know distribution of ST under Q. 19 / 46 Review: discounted price process is martingale under Q Remember from the discrete-time dynamic model, we know that for all t i h S i h St T Q ST St = EQ ⇒ = E . t t Rft RfT RfT −t In our new notation, writing βt = Rft we have hS i St T = EQ . t βt βT In other words, S/β is a martingale under Q. Note: The above ignores dividends, but see HW3. In general, discounted value process (i.e., with dividends reinvested) is martingale under Q. 20 / 46 Plan 1. Define Q such that discounted price (or value) process is martingale. ⇒ Under Q, drift of St /βt (or Vt /βt if dividends) is 0. ⇒ Under Q, suppose BtQ is a BM, so that for some αt we have d(St /βt ) = αt dBtQ . 2. Then we can write SDE for S in terms of Q. 3. Integrate this SDE to find the distribution of ST under Q. 4. Use distribution to compute option price from h i + . Ct = e−r(T −t) EQ (S − K) T t 21 / 46 Girsanov’s Theorem (how to find Q and BtQ ) Theorem Assume {Bt }0≤t≤T is a BM on probability space (Ω, F, P ), {Ft }0≤t≤T is a filtration for this BM, and {ηt }0≤t≤T is an adapted process Define Zt = exp n Z − 0 t 1 ηs dBs − 2 Z t o ηs2 ds , 0 RT and assume that EP [ 0 ηs2 Zs2 ds] < ∞. Then EP [ZT ] = 1, and under probability measure Q defined by EQ [X] = EP [ZT X], for all random variables X, Rt the process BtQ = Bt + 0 ηs ds is a BM on probability space (Ω, F, Q). 22 / 46 Finding risk-neutral measure - general setup Risk-free: Consider an adapted interest rate process rt , and define compounded return process (money market value): Rt βt = e Let Xt = Rt 0 0 rs ds . rs ds, so βt = eXt . By Itô’s lemma, dβt = rt βt dt. Stock: Assume stock price follows generalized geometric BM, i.e., it satisfies the SDE dSt = µt St dt + σt St dBt , ∀0 ≤ t ≤ T . 23 / 46 Finding risk-neutral measure - discounted price process The discounted stock price process is St /βt . Let f (β, s) = βs . Use two-dimensional Itô’s lemma: d(St /βt ) = fs dSt + fβ dβt 1 + fss Vart [dSt ] + fββ Vart [dβt ] + 2fsβ Covt [dSt , dβt ] 2 Note that fs = 1/β, fss = 0, fβ = −s/β 2 , Vart [dβt ] = 0, and Covt [dSt , dβt ] = 0, so St St dSt St 1 − rt dt. d(St /βt ) = dSt − 2 dβt = βt βt St βt βt Thus, after substituting form of dSt , d (St /βt ) dSt = − rt dt = (µt − rt )dt + σt dBt St /βt St Define the market price of risk ηt := µt −rt σt , and write in terms of ηt : d (St /βt ) = σt [ηt dt + dBt ]. St /βt 24 / 46 Finding risk-neutral measure - using Girsanov With Girsanov’s theorem, if we use Z n Z t 1 t 2 o Zt = exp − ηs dBs − η ds 2 0 s 0 to define Q, then the process dBtQ = ηt dt + dBt is BM under Q. Then, under Q the discounted stock dynamics are d (St /βt ) = σt [ηt dt + dBt ] = σt dBtQ . St /βt Notice that St /βt is a Q-martingale (since after integrating, the right-hand-side is Itô integral with respect to a BM). Finally, recall dSt St = rt dt + dSt St d(St /βt ) St /βt , = so we have rt dt + σt dBtQ | {z } same vol σt , changes drift from µt to rt 25 / 46 Risk-neutral measure - discounted portfolio value process The discounted value of a portfolio, Vt /βt , is also martingale under Q. To see this, write the self-financing condition dVt = ∆t dSt + rt (Vt − ∆t St ) dt and use Itô’s lemma with the P -SDEs for βt and St /βt to find (similar to step 3 on slide 8): d (Vt /βt ) = ∆t d (St /βt ) . Using d(St /βt ) St /βt = σt dBtQ , we have d (Vt /βt ) = ∆t σt (St /βt )dBtQ , which is an Itô integral with respect to BM, so martingale. 26 / 46 Risk-neutral pricing Let CT be random variable representing derivative payoff at T . We seek initial capital V0 and portfolio process ∆t that replicates payoff, in particular VT = CT . Vt /βt is Q-martingale ⇒ under the ∆t that we are looking for, we have: hV i h i Vt T Q CT = EQ = E . t t βt βT βT But V is replicating portfolio value. By no-arbitrage, derivative’s price is Ct = Vt , so h RT i − t rs ds Ct = EQ e C T . t 27 / 46 Simple example - pricing a log contract Assume dSt St = µdt + σdBtP . Find price of derivative paying log ST at T . Girsanov Theorem: dSt = rdt + σdBtQ . St Itô’s lemma: 1 d log St = r − σ 2 dt + σdBtQ . 2 Integrate SDE ⇒ conditional on time-t information, 1 Q log ST ∼ Normal log St + r − σ 2 (T − t), σ 2 (T − t) . 2 Risk-neutral pricing ⇒ price of log contract is Lt = e−r(T −t) EQ t [log ST ] h i 1 = e−r(T −t) log St + r − σ 2 (T − t) . 2 28 / 46 Example - pricing an exotic derivative A bank sells derivative paying ( (log (ST ))2 + ST − K (log (ST ))2 if ST > K if ST ≤ K, where ST is stock price at expiration and K is strike. Assumptions: Stock pays no dividends. Stock follows geometric BM under true measure P , with µ = 0.1 and σ = 0.2. Initial stock price is S0 = 100. Risk-free rate is r = 0.05. Derivative expires in half a year, and has strike price K = 100. Find no-arbitrage price of derivative. 29 / 46 Example - pricing an exotic derivative - solution plan a. Write down SDE for St , using B P , the BM under true measure. b. Rewrite SDE using B Q , the BM under risk-neutral measure. c. Use Itô’s lemma to find SDE for log St under risk-neutral measure. d. Integrate SDE to find risk-neutral distribution of log ST at t = 0. e. Express the derivative’s price (at t = 0) as a risk-neutral expectation. f. Calculate the derivative’s price. 30 / 46 Example - pricing an exotic derivative - solution (a-c) a. SDE for S under true measure is dSt = 0.1dt + 0.2dBtP . St b. Girsanov’s theorem ⇒ can change drift to r = 0.05, so under risk-neutral measure Q, we have dSt = 0.05dt + 0.2dBtQ . St c. We have log St = f (St ) with f (x) = log x, so Itô’s lemma yields 1 d log St = df (St ) = f 0 (St )dSt + f 00 (St )Vart [dSt ] 2 1 1 1 = dSt − (0.04St2 dt) St 2 St2 = 0.03dt + 0.2dBtQ . 31 / 46 Example - pricing an exotic derivative - solution (d-e) d. Integrating the SDE we have Z log ST = log S0 + T Z 0.03dt + 0 0 T 0.2dBtQ = log S0 + 0.03T + 0.2(BTQ − B0Q ) Q = log(100) + 0.03(0.5) + 0.2B0.5 . Q Therefore, log ST ∼ Normal(4.62, 0.02). e. Derivative’s payoff is sum of two separate payoffs: (log ST )2 , which we always get, and ST − K, which we only get if ST > K. Price of derivative is sum of “price” of these separate payoffs: h i 2 + C0 = e−rT EQ (log S ) + (S − K) T T 0 h i h i 2 −rT Q log ST + = e−rT EQ (log S ) + e E (e − K) . T 0 0 {z } | Black-Scholes call price 32 / 46 Example - pricing an exotic derivative - solution (f) f. Continuing from above, we have: n h i h io 2 log ST C0 = e−rT EQ + EQ − K)+ 0 (log ST ) 0 (e m − log K o n 1 m + s2 − log K − KΦ , = e−rT v 2 + m2 + exp m + v 2 Φ 2 v v where m = 4.62 and v 2 = 0.02. Here we used fact that if X ∼ Normal(m, v 2 ) then 1 2 E[max(eX − K, 0)] = em+ 2 v Φ m + v 2 − log K v − KΦ m − log K v Plugging in the numbers, we have: n o 0.02 √ √ C0 = 0.02 + 4.622 + e4.62+ 2 Φ 4.62+0.02−4.605 − 100Φ 4.62−4.605 e−0.05×0.5 0.02 0.02 = 27.7. 33 / 46 Risk-neutral pricing works for very general payoffs As the previous example illustrates, it doesn’t really matter what the final payoff looks like. If X is the underlying, and g(XT ) is the final payoff for some arbitrary function g, then the time-t price is h RT i − t ru du Pt = EQ e g(X ) T t TheR only challenge is trying to figure out the probability distribution of T e− t ru du g(XT ) under Q. Zero-coupon bond: g(x) = 1. European option: g(x) = max[x − k, 0] and XT = ST . If the underlying stock St is a GBM, then we showed the distribution of ST is lognormal, which makes this easy (Black-Scholes formula). If St is more complicated (e.g., stochastic volatility σt ), then ST will not be lognormal, so the solution won’t be as pretty. RT Asian options: g(x) = max[x − k, 0] and XT = T1 0 St dt. Barrier options: g(x) = max[x − k, 0] and XT = ST 1{inf t St >b} . 34 / 46 Risk-neutral pricing for dividend-paying stock Stock follows dSt St = µdt + σdBt and pays dividend δSt dt. Self-financing strategy: dVt = ∆t dSt + r (Vt − ∆t St ) dt + ∆t δSt dt. Using Itô’s lemma, we have Vt 1 dβt + dVt βt2 βt Vt 1 = − rdt + (∆t dSt + r (Vt − ∆t St ) dt + ∆t δSt dt) βt β h t i = ∆t (St /βt ) (µ − r + δ) dt + σdBt . d(Vt /βt ) = − Girsanov Theorem ⇒ for any ηt , there exists Q such that dBtQ = ηt dt + dBt is BM, so i St h d(Vt /βt ) = ∆t (µ − r + δ − σηt )dt + σdBtQ . βt Thus, Vt /βt is a Q-martingale (which is required under no-arbitrage) if and only if ηt = µ−r+δ , and so we have σ dSt /St = (r − δ)dt + σdBtQ . 35 / 46 Extra Topics – not for exam 36 / 46 Risk-neutral measure with multiple stocks Assume there are m stocks, each with SDE d X dSit = µit Sit dt + Sit σijt dBjt , i = 1, . . . , m. j=1 Rt As before, define compounded interest process βt = e d (Sit /βt ) = (µit − rt ) dt + Sit /βt d (Vt /βt ) = m X d X 0 rs ds , and find: σijt dBjt j=1 ∆it d (Sit /βt ) , i=1 where ∆it is the adapted portfolio process for each stock i. Definition A probability measure Q is risk-neutral if i) P and Q are equivalent (same null-sets), and ii) Under Q, the discounted stock price Sβitt is a martingale, ∀i. 37 / 46 Risk-neutral measure with multiple stocks To make the discounted prices martingales, we would like to find {ηjt } s.t. d d (Sit /βt ) X = σijt (ηjt dt + dBjt ) Sit /βt j=1 Then we could use Girsanov’s Theorem to construct the equivalent measure Q under which dBQ t = ηt dt + dBt would be a d-dimensional BM, and write d d (Sit /βt ) X Q σijt dBjt , = Sit /βt j=1 so discounted price would be sum of Q-martingales, hence Q-martingale. Clearly, to find {ηjt } we need to solve: ∀i : µit − rt = d X j=1 σijt ηjt ⇔ µt − rt 1 = Σt ηt | {z } in vector form 38 / 46 Market prices of risk - like state prices We want to solve µt − rt 1 = Σt ηt . |{z} |{z} |{z} m×1 m×d d×1 If m = d and Σt is full rank at all times, then the solution is ηt = Σ−1 t (µt − rt 1). Analogy to one-period model: with a complete market and no redundant assets, we solve for a unique state price vector. If m > d and Σt is full rank at all times, then the candidate solution (which may not satisfy the original equation) is ηt = (Σ0t Σt )−1 Σ0t (µt − rt 1). Analogy to one-period model: with redundant assets, there will exist a state price vector if and only if assets are priced correctly. If m < d and Σt is full rank at all times, then there are many solutions, one of which is ηt = Σ0t (Σt Σ0t )−1 (µt − rt 1). Analogy to one-period model: in an incomplete market with no redundant assets, there are many state price vectors. 39 / 46 American options Riskless rate is constant at r > 0 Remember – you never exercise the call early if r > 0 (HW3), so we focus on the put. Put price satisfies Pt = max τ ∈[t,T ] h i −r(τ −t) EQ max(K − Sτ , 0) t e {z } | present-discounted-value from exercise at τ | {z optimize over possible stopping times τ } Then, optimal exercise time is the first time exercising is at least as good as continuing (and exercising optimally thereafter), i.e., n o τ ∗ = inf τ : max(K − Sτ , 0) ≥ Pτ n o (by path-continuity) = inf τ : max(K − Sτ , 0) = Pτ 40 / 46 American options are difficult In general, it’s hard to price American options (and continuous-time doesn’t help that much). We know that the exercise rule n o τ ∗ = inf t : max(K − St , 0) ≥ Pt boils down to a time-dependent price threshold n o τ ∗ = inf t : St ≤ s∗ (t) for some function s∗ (t) (the exercise boundary). But what does s∗ (t) look like? Hard in general. 41 / 46 Perpetual American options Simpler setting: Suppose no expiration date, i.e., T → +∞ Also, suppose dSt = St [µdt + σdBt ], with S0 = s as in Black-Scholes. Result: the exercise boundary s∗ (t) no longer depends on time t And for the same reason, the put price no longer depends on time t, only the current stock price, so there is some function p such that Pt = p(St ) We want to find time-invariant exercise boundary. To do that, let s∗ < K be some arbitrary exercise boundary. Let τ be the first time t that St hits s∗ (or if St never hits s∗ , then we set τ = ∞ and the option is never exercised). If s∗ describes our exercise strategy, then the put value is −rτ max(K − Sτ , 0)1{τ <∞} ] p(s; s∗ ) = EQ 0 [e −rτ = (K − s∗ )EQ 1{τ <∞} ] 0 [e 42 / 46 Perpetual American options, continued −rτ 1 Need to compute EQ {τ <∞} ] 0 [e 1. Use Girsanov’s theorem: dSt = St [rdt + σdBtQ ] 2. Use Itô and integrate to get the usual St = s exp[(r − 21 σ 2 )t + σBtQ ] 3. At t = τ , we have St = s∗ , so rearrange to get e−rτ = 1 4. Plug this in, and notice that Zt := e− 2 σ martingale from Girsanov’s theorem, so −rτ EQ 1{τ <∞} ] = 0 [e 2 t+σB Q t s − 12 σ 2 t+σBtQ s∗ e is the same type of s Q s s E0 [Zt 1{τ <∞} ] = ∗ E∗0 [1{τ <∞} ] = ∗ P∗0 [τ < ∞], ∗ s s s where E∗ is a new expectation operator (and P∗ its associated probability) under which Bt∗ := BtQ − σt is a BM 5. Using the result of step 2 again, then step 4, note that P∗0 [τ < ∞] = P∗0 [(r + 12 σ 2 )t + σBt∗ ≤ log(s∗ /s) for some t > 0] 43 / 46 Perpetual American options, continued To solve this probability, we use the following fact, which can be proved using Doob’s optional stopping theorem, with the smart choice of martingale Mt := exp[−2a(at + Bt )] and clever choice of stopping time τ := inf{t : Mt = e2ab or Mt = 0}: Lemma If B is a BM under P, and if a > 0 and b < 0, then P0 [at + Bt ≤ b for some t > 0] = e2ab . Using this lemma, with a = r/σ + 21 σ and b = (1/σ) log(s∗ /s), we get 1 P∗0 [τ < ∞] = P∗0 [(r + σ 2 )t + σBt∗ ≤ log(s∗ /s) for some t > 0] 2 h r s∗ i s∗ 1+2r/σ2 1 = exp 2 2 + log = σ 2 s s Thus, s∗ 2r/σ2 −rτ EQ [e 1 ] = {τ <∞} 0 s 44 / 46 Perpetual American options, finish up Plug in the previous results into the put price to get s∗ 2r/σ2 p(s; s∗ ) = (K − s∗ ) s This holds for arbitrary choice of s∗ . To find the optimal exercise, we solve maxs∗ p(s; s∗ ). The FOC is s∗ 2r/σ2 2r K − s∗ s∗ 2r/σ2 − + = 0. s σ2 s∗ s Solve for s∗ to get 2r s∗ = K. 2r + σ 2 as σ increases, s∗ decreases [i.e., wait longer to exercise with more uncertainty, because more bad things can happen to the stock] as r decreases, s∗ decreases [i.e., wait longer to exercise under lower interest rates, because the opportunity cost of waiting declines] as r & 0, s∗ & 0 [i.e., never exercise with zero interest rates, as you showed in HW3] 45 / 46 The End! 46 / 46