Derivative Pricing and Financial Modelling Wei Wei Department of Actuarial Mathematics and Statistics Heriot-Watt University January 29, 2022 1 Introduction to financial derivatives A financial derivative is a contract whose value is based on the performance of an underlying asset. The underlying assets can be stock, commodities (metals, oil, other physical products), currencies, etc. According to the flexibility of execution, financial derivatives can be generally divided into two categories: European derivatives and American derivatives. A European derivative is a contract which limits execution on its maturity date. An American derivative is a contract which delivers its payoff at any time before or upon its maturity date. This course is about how to price a derivative at any time before the maturity date. The key to price a financial derivative is to tame the uncertainty of the derivative. Since the uncertainty arises from its underlying asset(s), the key to pricing derivatives is to use the underlying asset(s) to remove the risk from trading the derivative. The act of removing the uncertainty is called hedging. 1.1 Value of money Money has time value, that is, money available at the present time is worth more than the identical sum in the future. The time value of money is charactered by interest rates. There are several methods to calculating the value of money at different dates. Here we briefly introduce simple interest, compound interest and continuous interest. Suppose that the interest accumulation period is given by n, the interest rate in each period is given by r and the principle is given by P . Then under the simple interest accumulation rule, the value of the principle in the nth period is given by P (1 + nr). Under the compound interest rule, the value of the principle in the nth period is given by P (1 + r)n . In this course, we use the “continuous interest accumulation rule”, which is an extension of the compound interest rule, to compute the value of money. That is, the increase of money during [t, t + dt] is given by rMt dt, where r is the interest rate 1 per time unit and Mt is the value of money at time t. On the other words, the value of money evolves according to the differential equation given by dMt = rMt . dt Equivalently, for the principle M0 , the money value of money at time t is given by Mt = M0 ert . 1.2 远 期 Forwards A forward contract is an agreement to buy or sell an asset with a certain price at a certain date. A forward contract is usually traded in the over-the-counter (OTC) market. There are two parties in a forward contract. The one- who agrees to buy (sell) the underlying asset is said to hold a long (short) position. Suppose that in a forward contract an agent who takes a long position will buy the underlying asset, which evolves according to the process S = {St }tØ0 , with price K, which is called deliver price, at time T , which is called maturity, then the forward contract’s payoff at the maturity is given by 多空 场外交易 ST ≠ K. Similarly, from the perspective of a short position, the payoff of the forward contract at the maturity described above is given by K ≠ ST . 1.3 Futures A futures contract is a standardised forward contract, traded on an organised exchange,所 and such that, if a contract is traded at some time, the delivery price is ⼀ set to a special value Ft,T , called the futures price of the asset or the forward price of the asset, chosen so that the value of the futures contract at initiation (that is, at time t), zero. The two parties in a future contract should have faced a huge default risk, as they usually do not know each other. To guarantee the contract is honoured, a participant in a futures market is required to set up a so-called margin保证⾦ 兑现 account as collateral, and one’s daily profits and losses are reflected by adjustments 抵押品 in the account (the holder of a futures contract receives the change in value of the futures price after each day, for each contract held). One also has to maintain the balance in the margin account at some minimum value (the maintenance margin), and receives a so-called margin call (a demand to top-up the margin account) if the balance in the margin account falls below the maintenance margin. This mechanism is designed to remove the default risk险 from the market, and hence futures markets机制 are very liquid. Futures markets have other specialised features that we will not go further too much in this course. See Hull (2012) for more details. 贅 趟 inn 2 期权 1.4 Options An option contract is an agreement that gives the buyer of the contract a right (but not an obligation) to buy or sell an asset with a certain price at a certain date. The specified price is the strike price or exercise price of the option and the specified date is the maturity or expiration date of the option. There are two types of options in terms of buying or selling the underlying asset. A call option is a right to buy the underlying asset with the strike price at the maturity and a put option gives the buyer the right to sell the underlying asset with the strike price at the maturity. The payoff of a call option is given by (ST ≠ K)+ , where ST is the price of the underlying asset at the maturity T and K is the strike price. Similarly, the payoff function of a put option is given by (K ≠ ST )+ In terms of the flexibility of exercising the option, options can be categorised into two types. A buyer of a European option can exercise the right only at the maturity, while a buyer of an American option can exercise the right at or prior to the maturity. 3 2 Arbitrage-free principle In this section, we introduce the arbitrage-free principle, which is the foundation of derivative pricing and valuation. Then we illustrate, using the arbitrage-free principle, several properties of option price. 2.1 Arbitrage-free principle Consider a financial market, which comprises n+1 financial securities, (S0 , S1 , ..., Sn ). Here S0 is a risk-free asset (moeny account) and evolves according to dS0t = rS0t , dt with r > 0. (S1 , ..., Sn ) represents risky assets (e.g., stocks, index...), which have random returns on probability space ( , F, P). A wealth process is given by a portfolio process („0t , „1t , ..., „nt ) and the underlying assets in the market, i.e., t n ÿ = „it Sit . i=0 提取 注⼊ If during the entire period [0, T ] the investor does not infuse or withdraw fund in the wealth, then the portfolio is said to be self-financing. ⾃筹 资⾦ Definition 2.1. A self-financing portfolio is said to have an arbitrage opportunity in [0, T ], if there exits tú œ [0, T ), such that the wealth process tú Ø0 T P( =0 free > 0) > 0. T Take jake Definition 2.2. If there exists no arbitrage opportunity for any self-financing portfolio in [0, T ], then the market is said to be arbitrage-free in the period [0, T ]. Theorem 2.3. If the market is arbitrage-free in [0, T ], and processes, such that 1T and P( 1T Ø then for any t œ [0, T ), 1t > 4 2t . and 2 are wealth (1) 2T 2T ) > 1 > 0, (2) (3) Proof. Suppose that such that 1t 2t , Æ c t at time t œ [0, T ). Construct a portfolio process, = 1t ≠ 2t + M = 0, where M is a money account. Then at time T , we have that c T = 1T ≠ 2T + M er(T ≠t) Ø 1T ≠ 2T . Then it follows from (1) and (2) that the portfolio constructing c yields an arbitrage opportunity. This is a contradiction and thus completing the proof. Corollary 2.4. If the market is arbitrage-free in [0, T ] and 2t , ’t œ [0, T ]. 1T = 2T , then 1t = Proof. Consider a wealth process constructed by c = 1 ≠ 2 ⼆个正⽐⼀正it tMert + ‘M where ‘ > 0, M is a bank account. Then at time T , c T = ‘M erT > 0. 1t ≠ 正⽐如 不0 Then it follows Theorem 2.3 that c t = + ‘M ert Ø 0. 2t Let ‘ æ 0, then we have 1t ≠ Using the same method, we also prove 2.2 2t 2t Ø 0. ≠ 1t Ø 0. Therefore, 1t = 2t . European options and call-put parity In this section, we consider the relationship between the prices of call and put options. We suppose that the market is arbitrage-free and no transaction cost. Also, we suppose that the underlying stock pays no dividends. We define the underlying stock process by St , the European call option price by ct , European put option price by pt , the strike price by K, the expiration date by T and the constant risk-free interest rate by r. 股息 Theorem 2.5. Suppose that P(ST ≠ K > 0) > 0 and P(ST ≠ K < 0) > 01 , then (a) 1 (St ≠ Ke≠r(T ≠t) )+ < ct < St , ’t œ [0, T ). Without any specification, this assumption will be added throughout the course. 5 Ct St ru q 19 K⼋ STST PG k.to a so T 1 71 Ctc St 1 知 51 1 9 9 0inch 11 Theorem 23 (b) (Ke≠r(T ≠t) ≠ St )+ < pt < Ke≠r(T ≠t) , ’t œ [0, T ). Proof. We only prove part (a) and the proof of part (b) is similar. At t = 0, we construct a portfolio, such that 10 = S0 ≠ Ke≠rT and 20 = c0 . Then at time T , we have that 1T = ST ≠ K Æ (ST ≠ K)+ = 2T . As P(ST ≠ K < 0) > 0, then P( 1T < 2T ) > 0. Therefore, Theorem 2.3 yields that ct > St ≠ Ke≠r(T ≠t) , ’t œ [0, T ). Similarly, since cT = (ST ≠ K)+ and P(ST ≠ K > 0) > 0 , then it follows from Definition 2.1 that ct > 0.As a result, we have (St ≠ Ke≠r(T ≠t) )+ < ct . It is now suffices to prove ct < St , which follows from cT Æ ST . Theorem 2.6 (Call-put parity). ct + Ke≠r(T ≠t) = pt + St , ’t œ [0, T ]. (4) Proof. Consider two portfolios, such that 1 = c + Ke≠rT , 2 = p + S. At t = T, we have 1T = (ST ≠K)+ +K = max(K, ST ) and 2T = (K ≠ST )+ +ST = max(K, ST ). Therefore, 1T = 2T . Then Corollary 2.4 yields 1t = 2t , ’t œ [0, T ]. 2.3 American options and early exercises In contrast to a European option holder, a holder of an American option has a right to exercise the option early. Because of the early exercise right, it is easy to see that an American option has higher value than the European option with the same strike price, underlying asset and maturity. Define the American option call and put price at time t by Ct and Pt respectively. Therefore, Ct Ø ct , Pt Ø pt . The following theorem illustrates that the early exercise right is of no use for an American call option with a non-dividend paying underlying stock. Theorem 2.7. If a stock has no dividend, then the American call option with the stock being the underlying asset follows Ct = ct . Proof. It follows from Theorem 2.5 that Ct Ø ct > (St ≠ Ke≠r(T ≠t) )+ Ø (St ≠ K)+ , ’t œ [0, T ). Then it is to see that early exercise for the American option is unwise. 6 A natural question here is that: is the early exercise right for an American put option useful? The answer is yes. In fact, suppose that the underlying asset falls ⼀⼀ below K(1 ≠ e≠r(T ≠t) ) and the price of the American option is the same as the European option. Then Theorem 2.5 yields that nF stckcte n ke Pt = pt Æ Ke≠r(T ≠t) < K ≠ St . This contradicts Theorem 2.5. Theorem 2.8. If C and P are the prices of American call option and put option with a non-dividend underlying stock, and then St ≠ K < Ct ≠ Pt < St ≠ Ke≠r(T ≠t) , ’t œ [0, T ). Proof. We prove the left hand-side of the inequality. The right hand-side is similar and thus we omit it. At time t œ [0, T ), consider two portfolios, such that 1t = Pt + St , 2t If the American option is not early exercised, then 2T + = (ST ≠ K) + Ke 1T r(T ≠t) c SK = K + Ct . = max(ST , K) + K(er(T ≠t) ≠ 1), = (K ≠ ST )+ + ST = max(ST , K). Therefore 2T > 1T , which implies 2t > 1t . If the American option is early exercised at time · (t < · < T ) 2· = C· + Ker(· ≠t) , 1· t = max(S· , K). As a result, it follows from Theorem 2.5 that 2· This completes the proof. 2.4 > (S· ≠ K)+ + Ker(· ≠t) > Kerli 1· . Option price and strike price In this section, we study the dependence of option price on its strike price. For sake of simplicity, we suppose that the underlying assets in this section are non-dividend paying. Theorem 2.9. Let ct (K) be the price of a European call option with the strike price K. For two European call options c(K1 ), c(K2 ) with the same maturity date, if K1 > K2 , then 0 Æ ct (K2 ) ≠ ct (K1 ) Æ K1 ≠ K2 , ’t œ [0, T ]. 7 (5) Proof. We will prove the right side of inequality (2.5). The left side is similar and hence we omit the proof. Construct two portfolios, such that 1 = c(K1 ) + K1 , 2 = c(K2 ) + K2 . On the maturity date t = T, 1T = cT (K1 ) + K1 er(T ≠t) = (ST ≠ K1 )+ + K1 er(T ≠t) , 2T = cT (K2 ) + K2 er(T ≠t) = (ST ≠ K2 )+ + K2 er(T ≠t) . • If ST > K1 . then 1T • If ST < K2 , then 1T = ST + K1 (er(T ≠t) ≠ 1) > ST + K2 (er(T ≠t) ≠ 1) = 2T . • If K2 < ST < K1 , then 1T = K1 er(T ≠t) , 2T = ST +K2 (er(T ≠t) ≠1). Therefore r(T ≠t) + K2 ≠ ST > (K1 ≠ K2 )(er(T ≠t) ≠ 1) > 0. 1T ≠ 2T = (K1 ≠ K2 )e = K1 er(T ≠t) > K2 er(T ≠t) = 2T . Then Theorem 2.3 yields the result. Theorem 2.10. For two European put options with the same maturity date, if K1 > K2 , then 0 Æ pt (K1 ) ≠ pt (K2 ) Æ K1 ≠ K2 . 盥 凸 Theorem 2.11. European call (put) option price ct (K)(pt (K)) is a convex function of K, i.e., for K1 > K2 and K0 = ⁄K1 + (1 ≠ ⁄)K2 , (0 Æ ⁄ Æ 1). Then ct (K0 ) Æ ⁄ct (K1 ) + (1 ≠ ⁄)ct (K2 ), pt (K0 ) Æ ⁄pt (K1 ) + (1 ≠ ⁄)pt (K2 ). Proof. We prove the convexity of call options. The convexity of put options is similar hence the proof is omitted. Consider two portfolios, such that 1 Then 1T = c(K0 ), = (ST ≠ K0 )+ , 2T 2 = ⁄c(K1 ) + (1 ≠ ⁄)c(K2 ). = ⁄(ST ≠ K1 )+ + (1 ≠ ⁄)(ST ≠ K2 )+ Then the convexity of the call option price follows from the convexity of the payoff function (S ≠ K)+ with respect to K. 8 3 The Black-Scholes-Merton model In this section we introduce the Black-Scholes-Merton model. 3.1 Assumptions • The underlying asset follows from a geometric Brownian motion, i.e., dSt = µdt + ‡dWt St (6) where W = {Wt }tØ0 is a standard Brownian motion. µ and ‡ are constant and represent the expected return rate and volatility of the underlying asset respectively. 预期回报率 • The risk-free interest rate r is constant, • The underlying asset pays no dividend, • No transaction cost and tax. • The market is arbitrage-free. 3.2 3.2.1 Black-Scholes equation Delta hedging We introduce the celebrated Black-Scholes’ delta hedging rule (Black and Scholes 1973), which is based on the arbitrage-principle in the market (see Merton et al. 1973). The uncertainty of the price of options come from the randomness of the underlying asset. The delta hedging is to use the underlying asset to remove the uncertainty of the option price. Suppose the value of the option is defined by V . Construct a portfolio at time t œ [0, T ), t = Vt ≠ t St Here denotes the shares of the underlying asset. Choose Holding t during [t, t + ”t], we have t+”t ≠ t = Vt+”t ≠ Vt ≠ t (St+”t such that is risk-free. ≠ St ) Letting ”t æ 0, then we have that d t =r t dt = r(Vt ≠ 9 t St )dt (7) dVt ≠ ˆV 1 ˆ2V ˆV (t, St ) + ‡ 2 St2 2 (t, St ) + µSt (t, St ))dt ˆt 2 ˆS ˆS ˆV + ‡St (t, St )dWt ≠ t (µSt dt + ‡St dWt ) ˆS t dSt = ( To remove the uncertainty, we let ( t = ˆV (t, St ), ˆS (8) (9) then (7) and (8) yield that ˆV 1 ˆ2V ˆV (t, St ) + ‡ 2 St2 2 (t, St ) + µSt (t, St ))dt ≠ ˆt 2 ˆS ˆS t µSt dt = r(Vt ≠ t St )dt which implies ˆV 1 2 2 ˆ2V ˆV (t, S) + ‡ S (t, S) + rS (t, St ) ≠ rV = 0, 2 ˆt 2 ˆS ˆS (10) for ’(t, S) œ [0, T ) ◊ (0, Œ). Moreover, at t = T , the payoff function yields that V (T, S) = I (S ≠ K)+ (K ≠ S)+ (call option) (put option). (11) Equation (10) shows the dynamics of a European contingent. This equation is first established by Black and Scholes (1973) and thus called the Black-Scholes (BS) equation. A distinctive feature of BS equation is that it does not depend on the drift of the underlying asset. This feature sparks the so-called risk neutral pricing, which opens a new way to study the derivative pricing theory and the financial market. 3.2.2 Replication Another way to derive the Black-Scholes equation is replication, that is, replicating the option by constructing a portfolio comprising the underlying asset and the riskfree asset. We now try to replicate the option by a portfolio, which means we are looking for a self-financing portfolio process („0 , „1 ), such that Vt = „0t Mt + „1t St , (12) and d(„0t Mt + „1t St ) = „0t dMt + „1t dSt , or equivalently Mt d„0t + d„0t dMt + St d„1t + dSt d„1t = 0, where Mt is the risk free asset2 . 2 Condition (13) is called self-financing condition. 10 (13) Consider the dynamics of Vt , then we have ˆV 1 ˆ2V ˆV (t, St ) + ‡ 2 St2 2 (t, St ) + µSt (t, St ))dt ˆt 2 ˆS ˆS ˆV + ‡St (t, St )dWt ˆS dVt = ( (14) (15) On the other hand, the dynamics of „0t Mt + „1t St is given by d(„0t Mt + „1t St ) = „1t (µSt dt + ‡St dWt ) + „0t rMt dt Then it follows from (12) that d(„0t Mt + „1t St ) = „1t (µSt dt + ‡St dWt ) + r(Vt ≠ „1t St )dt (16) Choosing „1t = ˆV (t, St ) and building up the equation between (14) and (16), we ˆS obtain the Black-Scholes equation (10). 3.3 The Black-Scholes Formula We solve the BS equation (10) with the terminal condition (11) in this section. We consider the call option case, the put option case is similar and we thus omit it. Let x = ln S, · = T ≠ t, then (10) becomes ˆV 1 2 ˆ2V 1 ˆV ≠ + ‡ + (r ≠ ‡ 2 ) ≠ rV = 0, 2 ˆ· 2 ˆx 2 ˆx (·, x) œ (0, T ] ◊ (≠Œ, Œ) (17) with the initial condition V (0, x) = (ex ≠ K)+ , x œ (≠Œ, Œ). (18) Equation (17) and initial condition (3.3) are called Cauchy problem in partial differential equation (PDE) theory. To solve the Cauchy problem, we set V = ue–· +—x , (19) where –, — are chosen so that equation (17) is reduced to a heat equation. By simple calculation, we have V· = e–· +—x (u· + –u) Vx = e–· +—x (ux + —u) Vxx = e–· +—x (uxx + 2—ux + — 2 u). 11 Substitute the above three equations into (17) and eliminate e–· +—x , then we have 1 1 ‡2 ‡2 u· ≠ ‡ 2 uxx ≠ (—‡ 2 + r ≠ ‡ 2 )ux + (r ≠ —(r ≠ ) ≠ — + –)u = 0 2 2 2 2 We choose 1 r ≠ 2, 2 ‡ 2 ‡ ‡2 1 ‡2 – = ≠r + —(r ≠ ) + — 2 = ≠r ≠ 2 (r ≠ )2 . 2 2 2‡ 2 The by transformation (19), the equation (17) is reduced to —= 1 u· ≠ ‡ 2 uxx = 0 2 (20) u(0, x) = e≠—x (ex ≠ K)+ (21) with the initial condition As we know the solution of the Cauchy problem corresponding to the heat equation (20) is given by the Poisson formula u(·, x) = ⁄ Œ ≠Œ K(x ≠ ›, · )„(›)d›, where „(›) is the initial value, K(x ≠ ›, · ) is the fundamental solution of the heat equation (20): (x≠›)2 1 K(x ≠ ›) = Ô e≠ 2‡2 · . ‡ 2fi· Therefore, the solution u can be written as ⁄ Œ (x≠›)2 1 Ô e≠ 2‡2 · e≠—x (ex ≠ K)+ d› ≠Œ ‡ 2fi· ⁄ Œ (x≠›)2 1 Ô = e≠ 2‡2 · e≠—x (ex ≠ K)d› ln K ‡ 2fi· u(·, x) = 2 Note that — = 12 ≠ ‡r2 , and – = ≠r ≠ 2‡1 2 (r ≠ ‡2 )2 . Then back to the original function V (·, x), we have 1 1 V (·, x) = e( 2 ≠ ‡2 )x+(≠r≠ 2‡2 (r≠ r ‡2 2 ) )· 2 u(·, x) = I1 + I2 , where I1 = e ≠r· ⁄ Œ ln K 2 (x≠›+(r≠ ‡2 1 2‡ 2 · Ô e≠ ‡ 2fi· 12 )· )2 +› d›. Denote ÷ = x ≠ › + (r ≠ then ‡2 )·, 2 ‡2 ex ⁄ x≠ln K+(r≠ 2 )· ≠ (÷+‡22 · )2 I1 = Ô e 2‡ · d÷. ‡ 2fi· ≠Œ Define Ê = 2 ÷+‡ Ô · , N (x) ‡ · = Ô1 2fi sx Ê2 ≠ 2 dÊ, then ≠Œ e x ≠ ln K + (r + Ô I1 = e N ( ‡ · x ‡2 )· 2 ). Similarly, we have that I2 = ≠Ke ≠r· x ≠ ln K + (r ≠ Ô N( ‡ · ‡2 )· 2 ). Back to the original variables (t, S), we have V (t, S) = SN (d1 ) ≠ Ke≠r(T ≠t) N (d2 ), (22) where 2 S ln K + (r + ‡2 )(T ≠ t) Ô d1 = ‡ T ≠t 2 ln S + (r ≠ ‡2 )(T ≠ t) Ô d2 = K ‡ T ≠t Equation (22) is the celebrated Black-Scholes formula for a call option. Moreover, by the call-put parity (4), we obtain the value of the put option p(t, S) = c(t, S) + Ke≠r(T ≠t) ≠ S = Ke≠r(T ≠t) N (≠d2 ) ≠ SN (≠d1 ). (23) Equation (23) is the Black-Scholes formula for a put option. 3.4 Derivation of the Black-Scholes formula: a probabilistic perspective In this subsection, we derive the Black-Scholes formula using the Feynman-Kac formula, which provides a linkage between PDEs and conditional expectations. Let us focus on the European call option. The European put option price can be obtained then by the call-put parity. It follows from the Feynman-Kac formula that 13 the solution to the PDE problem (10) and (11) can be represented by the following conditional expectation c(t, S) = E[e≠r(T ≠t) (XT ≠ K)+ |Xt = S], where dXt = rXt dt + ‡Xt dWt . Solving the SDE, we have that 1 XT = S exp((r ≠ ‡ 2 )(T ≠ t) + ‡(WT ≠ Wt )). 2 Let WT ≠ Wt Y =≠ Ô . T ≠t Then Ô 1 c(t, S) = E[e≠r(T ≠t) (S exp((r ≠ ‡ 2 )(T ≠ t) ≠ ‡Y T ≠ t) ≠ K)+ ]. 2 It is easy to see that Ô Y follows the standard normal distribution and S exp((r ≠ 1 2 ‡ )(T ≠ t) ≠ ‡Y T ≠ t) ≠ K > 0 if and only if y < d2 . Therefore, 2 Ô y2 1 ⁄ d2 ≠r(T ≠t) 1 c(t, S) = Ô e (S exp((r ≠ ‡ 2 )(T ≠ t) ≠ ‡y T ≠ t) ≠ K)e≠ 2 dy 2 2fi ≠Œ ⁄ d2 Ô 1 1 1 =Ô (S exp(≠ ‡ 2 (T ≠ t) ≠ ‡y T ≠ t ≠ y 2 ))dy ≠ e≠r(T ≠t) KN (d2 ) 2 2 2fi ≠Œ ⁄ d2 Ô S 1 =Ô exp(≠ (y + ‡ T ≠ t)2 )dy ≠ e≠r(T ≠t) KN (d2 ) 2 2fi ≠Œ ⁄ d2 +‡ÔT ≠t 1 2 S =Ô e≠ 2 z dz ≠ e≠r(T ≠t) KN (d2 ) 2fi ≠Œ = SN (d1 ) ≠ Ke≠r(T ≠t) N (d2 ). 3.5 Sensitivity parameters (Greeks) Delta. The derivatives of the call option pricing function c(t, s) with respect to various variables are called sensitivity parameters, or Greeks. A straightforward computation yields that t = ˆc (t, s) = N (d1 ) ˆs It is to see that the delta of a call option is positive. It means that if one sells a call option, it is hedged with a dynamically adjusted long position in the stock. 14 Similarly, the delta of a put option can be obtained by the call-put parity, which ˆc implies that ˆp = ˆx ≠ 1, then we have ˆs t = ˆp (t, s) = ≠N (≠d1 ). ˆs It is to see that the of a put option is negative. It means that if one sells a put option, it is hedged with a dynamically adjusted short position in the stock. Theta. The theta of a call option is ˆc ‡s = ≠rKe≠r(T ≠t) N (d2 ) ≠ Ô N Õ (d1 ). ˆt 2 T ≠t As N (·) and N Õ (·) are both positive, theta is always negative, meaning that the price of an option declines as we approach the maturity. This is true regardless of whether the option is a call or put, which you can easily verify using put-call parity. Gamma. The gamma of a call option is ˆ2c 1 Ô (t, s) = N Õ (d1 ), 2 ˆs ‡s T ≠ t 2 which is always positive and is equal to ˆˆs2p , the put gamma. Gamma is closely related to the risk introduced into the BS hedging program if trading is not continuous. Note that Gamma measures how quickly the delta of an ⼀ as the stock price changes. If the magnitude of gamma is small, then option changes delta changes slowly, so a trader will not have to re-hedge very often in order to maintain delta neutrality. On the other hand, if the magnitude of gamma is large, then the trader must re-hedge very often to maintain delta neutrality. Vega. The vega of an option is the derivative of the option price with respect to volatility, and given by Ô ˆc (t, s) = s T ≠ tN Õ (d1 ). ˆ‡ 对称 不 The positivity of vega shows the asymmetric change of the option price with respect to the volatility. Take a call option as an example. The holder benefits from the increase of the stock price, but has only limited downside risk in the decrease of the stock. Therefore the call option price increases as the volatility increases. 15 4 Risk neutral pricing It is worth noting that the return rate µ of the stock price does not appear in the BS equation. In this section, we introduce a probabilistic way to solve the BS equation based on this observation. 4.1 Two mathematical theorems Theorem 4.1 (Girsanov, one dimension). Let {Wt }tœ[0,T ] be a Brownian motion on a probability space ( , F, P), and let F = {Ft }tœ[0,T ] be a filtration for the Brownian motion. Let {◊t }tœ[0,T ] be an adapted process. Define Zt = e ≠ st 0 ◊s dWs ≠ 12 ⁄ t WtQ = Wt + 0 st 0 ◊s2 ds ◊s ds, and assume that 1 E[e 2 sT 0 ◊s2 ds ] < Œ. (24) Then E[ZT ] = 1 and under the probability measure Q given by Q(A) = ⁄ A ZT dP, the process W Q is a Brownian motion. Condition (24) is Novikov’s condition, which ensures that Z is a martingale under P. Theorem 4.2 (Martingale representation, one dimension). Let {Wt }tœ[0,T ] be a Brownian motion on a probability space ( , F, P), and filtration F = {Ft }tœ[0,T ] is generated by the Brownian motion. Let {Mt }tœ[0,T ] be a martingale with respect to this filtration. Then there is an adapted process { t }tœ[0,T ] , such that Mt = M0 + 4.2 ⁄ t 0 s dWs , ’t œ [0, T ]. Stock price under the risk neutral measure Let {Wt }tœ[0,T ] be a Brownian motion on a probability space ( , F, P), and let F = {Ft }tœ[0,T ] be a filtration for the Brownian motion. Consider the stock price governed by dSt = µt St dt + ‡t St dWt , 16 where µ and ‡ are allowed to be adapted processes. We assume that ‡t > 0, ’t œ [0, T ]. In addition, we suppose that the interest rate is an adapted process {rt }tœ[0,T ] . We define the discount process by Dt = e≠ Note that st 0 rs ds . dD(t)St = (µt ≠ rt )Dt St dt + ‡t Dt St dWt = ‡t St Dt (◊t dt + dWt ), where we define the market price of risk to be ◊t = µt ≠ r t . ‡t (25) We introduce the probability measure Q defined in Theorem 4.1, with the market price of risk ◊t given by (25). Then we rewrite the dynamics of the stock price as dDt St = Dt St ‡t dWtQ where W Q is define in Theorem 4.1 with the market price of risk ◊ given by (25). We call Q the risk neutral measure because it is equivalent to the original P and it renders the discounted stock price Dt St into a martingale. The undiscounted stock price St follows the dynamics dSt = rt St dt + ‡t St dWtQ . 4.3 (26) Portfolio process under the risk neutral measure In this section, we study portfolio processes under the risk neutral measure Q. Consider the portfolio comprising shares of stocks and a certain amount of riskfree asset. The the instantaneous change of the portfolio is given by dXt = t dSt + rt (Xt ≠ t St )dt = t (µt St dt + ‡t St dWt ) + rt (Xt ≠ = rt Xt dt + t St ‡t (◊t dt + dWt ). t St )dt (27) Then it follows from Ito’s formula that dDt Xt = t St ‡t (◊t dt + dWt ) = t d(Dt St ). Therefore, combined with the dynamics of the stock price (26), we have that dDt Xt = Q t ‡t Dt St dWt . 17 (28) 4.4 Pricing under the risk neutral measure and hedging In Section 3, we introduced the pricing mechanism based on the delta hedging strategy. If a contingent claim can be replicated by the underlying asset and the risk-free asset, then the price of the claim can be obtained by solving a PDE problem. The discussion in Section 3 depends on the Markov property of the payoff function. In this section, we extend the pricing mechanism to a more general case, which removes the Markov property of the payoff function. In section 3.2.2, we see that the key to price a contingent claim is to replicate the claim by the underlying asset and the risk-free asset. Following this idea, we first suppose that the contingent claim yields a payoff VT at the maturity T , where VT is FT measure. Note that we here do not suppose thatVT is a function of ST and thus not imposing the Markov property on the payoff function. Second, we suppose that the contingent claim can be replicated by the portfolio process X defined by (27). (Roughly speaking, if every contingent claim can be replicated by the stock and the risk free asset in the market, then we say the market is complete.) Suppose that there exists a portfolio process X, such that XT = VT Then it follows Corollary 2.4 that Xt = Vt . Moreover, the discounted portfolio process (28) tells that Dt Xt is a martingale under the risk neutral measure Q. Then Dt Vt = Dt Xt = EQ [DT XT |Ft ] = EQ [DT VT |Ft ], (29) which yields the risk neutral pricing formula for a European contingent claim Vt = 1 Q E [DT VT |Ft ] Dt (30) A natural question here is that: how can we find the hedging strategy ? We now answer this question by the martingale representation. Following (29), it is easy to see that Dt Vt is a martingale under the risk neutral measure Q. The the martingale representation theorem (Theorem 4.2) yields that Dt Vt = V0 + ⁄ t Q s dWs , 0 0 On the other hand, for any portfolio process is given by Dt Xt = X0 + ⁄ t 0 18 t, Æ t Æ T. the discounted wealth process X Q s ‡s Ds Ss dWs In order to have Xt = Vt , ’t œ [0, T ], we choose X0 = V0 and choose t such that t ‡t Dt St = t, which yields t 4.5 = t ‡t Dt St , ’t œ [0, T ], t œ [0, T ]. Forwards and futures In this subsection, we study the application of the risk neutral pricing introduced in subsection 4.4 on the valuation of forwards and futures. We assume that every contingent claim can be hedged and the underlying assets are non-dividend paying. We consider all problems in this subsection in the filtered probability space ( , F, {Ft }0ÆT̂ , P). Here, T̂ is a large number so that all financial securities expire before time T̂ . We suppose that there is a unique risk-neutral measure Q. Forwards. Let {St }tœ[0,T̂ ] be an asset price process and let {rt }tœ[0,T̂ ] be an st interest rate process. As usual, we define the discount factor by Dt = e 0 rs ds . It follows from the risk neutral pricing theory, the price at time t of a zero-coupon bond paying 1 at time T is given by ≠ B(t, T ) = 1 Q E [DT |Ft ]. Dt (31) Definition 4.3. A forwards contract is an agreement to pay a specified delivery price K at a delivery date T , for the asset price {St }tœ[0,T̂ ] . The T ≠forward price F orS (t, T ) of this asset at time t, where t œ [0, T ], is the value of K such that the forwards contract has no-arbitrage price zero at time t. Theorem 4.4. F orS (t, T ) = St , t œ [0, T ]. B(t, T ) Proof. It follows from the risk neutral pricing theory that, 0= 1 Q E [DT (ST ≠ K)|Ft ], Dt which implies 1 Q 1 Q E [DT ST |Ft ] = E [DT K|Ft ], Dt Dt 19 (32) which yields that St = KB(t, T ). This completes the proof. Futures. A futures contract is a standardised forward contract, traded on an organised exchange, and such that, if a contract is traded at some time, the delivery price is set to a special value Ft,T , called the futures price of the asset or the forward price of the asset, chosen so that the value of the futures contract at initiation (that is, at time t), zero. A distinctive feature of futures is the margin mechanism, which differs futures from forwards. In contrast to a forward contract which yields a lumpsum payoff at the maturity, a futures contract provides a cash flow. To see this, let us consider a time interval [0, T ], which we divide into subintervals using partition points 0 = t0 < t1 · · · < tn = T. We shall refer to each subinterval [tk .tk+1) as a day. Suppose that the interest rate is constant within each day. Then the discount process is given by D0 = 1 and, for k = 1, 2, ·, n ≠ 1, Dtk+1 = exp{≠ k ÿ j=0 rtj (tj+1 ≠ tj )}, which is Ftk ≠measurable. Consider a futures holder who takes a long position between time tk and tk+1 . At time tk , he receives Ftk+1 ,T ≠ Ftk ,T , which is called marking to margin. At each time tk , the future contract price equals zero, thus the risk neutral pricing theory yields that 1 Q E [Dtk+1 (Ftk+1 ,T ≠ Ftk ,T )|Ftk ] = 0. Dtk As Dtk+1 is Ftk ≠measurable, we have EQ [Ftk+1 ,T ≠ Ftk ,T |Ftk ] = 0, which implies that Ft,T , t = t0 , t1 , · · · , tn is a martingale under Q. Since FT,T = ST , then we have that Ft,T = EQ [ST |Ftk ], k = 0, 1, 2, · · · .n. We now extend this idea into the continuous time case. Definition 4.5. The futures price of an asset whose value at time T is ST is given by Ft,T = EQ [ST |Ft ], 20 t œ [0, T ]. Theorem 4.6. The futures price is a martingale under the risk neutral measure Q, it satisfies FT,T = ST , and the value of a long (or a short) futures position to be held over an interval of time is always zero. Proof. It is easy to see that Ft,T is a martingale under the risk neutral measure Q, it satisfies FT,T = ST . Therefore, it suffices to prove that the long or short position of a future contract values 0. Consider an agent who takes shares of future contract. Suppose the profit which the shares of future contract create is defined by X. Then we have that dXt = t dFt,T + rXt dt, which implies that dDt Xt = Dt t dFt,T This suggests that {Dt Xt }tœ[0,T ] is a martingale under Q, as Ft,T is a martingale under Q. Moreover, since holding any position ofs a futures does not trigger any payment, we have that X0 = 0. Therefore Dt Xt = 0t Ds s dFs,T , which implies that EQ [Dt Xt ] = 0. Let t = 1(≠1), ’t œ [0, T ], we have that the long (short) position of a future contract values 0. In the end of this subsection, we compare the forward and futures prices. If the interest rate process r is deterministic, then it is to see that 1 Q E DT [ST |Ft ] DT 1 St = Dt St = = F orS (t, T ). DT B(t, T ) Ft,T = EQ [ST |Ft ] = If r is a general stochastic process, then the forward-future spread is given by F orS (0, T ) ≠ Ft,T = = S0 Q E DT ≠ EQ [ST ] = 1 EQ DT 1 cov Q (DT , ST ), B(0, T ) [EQ [DT ST ] ≠ EQ DT EQ ST ] where cov Q is the covariance under the risk neutral measure Q. 4.6 4.6.1 Fundamental theorems of asset pricing Two mathematical theorems In this subsection, we study the financial market theory using risk neutral pricing theory. The mathematical tools are the multi-dimensional Girsanov theorem and the mult-dimensional martingale representation theorem. 21 In this subsection, let Wt = (W1t , W2t , · · · , Wnt ) be a n≠ dimensional Brownian motion3 defined on a probability space ( , F, {Ft }tØ0 , P), and Ft }tØ0 the filtration generated by W. Theorem 4.7 (Girsanov, multi-dimension). Let {(◊1t , ◊2t , · · · , ◊nt )}tœ[0,T ] be an adapted process. Define Zt = e≠ s t qn ◊ dWjs ≠ 12 j=1 js 0 WitQ = Wit + ⁄ t 0 st 0 ||◊||2s ds i = 1, 2, · · · , n, ◊is ds, and assume that E[e 1 2 sT 0 ||◊s ||2 ds ] < Œ. (33) Then E[ZT ] = 1 and under the probability measure Q given by Q(A) = ⁄ A ZT dP, the process W Q is a Brownian motion. Condition (33) is Novikov’s condition, which ensures that Z is a martingale. Theorem 4.8. [Martingale representation, multi-dimension] Let {Mt }tœ[0,T ] be a martingale with respect to {Ft }tœ[0,T ] . Then there is an adapted process {( 1t , 2t , · · · , such that Mt = M 0 + ⁄ tÿ n 0 j=1 js dWjs , ’t œ [0, T ]. We now consider the model of the market. Suppose that there are m stocks in the market, each with the dynamics given by n ÿ dSit = µit dt + ‡ijt dWjt , Sit j=1 i = 1, 2, · · · , m. where µi and Ò ‡ij are adapted processes. qn 2 Let ‡it = j=1 ‡ijt , which we assume is never 0. and we define Bit = ⁄ tÿ n 0 ‡ijs dWjs , j=1 ‡is 3 i = 1, 2, · · · , m. We say W is a n≠ dimensional Brownian motion, if (i) each Wi is one dimensional Brownian motion; (ii) if i ”= j, then Wi and Wj are independent. 22 nt )}tœ[0,T ] , It is easy to see that Bi is a continuous martingale and furthermore, 2 ‡ijt dBit dBit = dt = dt, 2 j=1 ‡i n ÿ i = 1, 2, · · · , m. Then it follows from Levy’s theorem that Bi is a Brownian motion. We may rewrite the stock price in terms of the Brownian motion Bi , dSit = µit Sit dt + ‡it Sit dBit , i = 1, 2, · · · , m. For i ”= k, Bi and Bk are typically not independent. To see this, let us consider dBit dBkt = flikt dt where flikt = qn ‡ijt ‡kjt , ‡it ‡kt j=1 We call flit the instantaneous correlation between Bi and Bk . Then Ito’s formula yields that Bit Bkt = which implies ⁄ t 0 ⁄ t Bis dBks + 0 Bks dBis + cov(Bit , Bkt ) = E Note that ⁄ t 0 ⁄ t 0 fliks ds, fliks ds. dSit dSkt = ‡it ‡kt Sit Skt dBit dBkt = flikt ‡it ‡kt Sit Skt dt which can be rewritten as dSit dSkt = flikt ‡it ‡kt dt. Sit Skt The volatility processes ‡i and ‡k are the respective instantaneous standard derivations of the relative changes in Si and Sk , and the process flik is the instantaneous correlation between these relative changes. Define a discount process Dt = e ≠ st 0 rs ds , where r is an adapted process. The discounted stock process is given by dDt Sit = Dt Sit ((µi ≠ rt )dt + ‡it dBit ), 23 i = 1, 2, · · · , m. (34) 4.6.2 Existence of the risk-neutral measure Definition 4.9. A probability measure Q is said to be risk neutral if (i) Q and P are equivalent (i.e., for every A œ F, P(A) = 0 if and only if Q(A) = 0), and (ii) under Q, the discounted stock price Dt Sit is a martingale for every i = 1, 2. · · · , m. In order to make discounted stock prices be martingales, we would like to rewrite the stock price as dDt Sit = Dt Sit n ÿ ‡ijt (◊jt + dWjt ), j=1 i = 1, 2, · · · , m. (35) If we can find (◊1 , ◊2 , · · · , ◊n ), such that (35) holds, then by the Girsanov theorem, we can construct a measure Q, which is risk neutral. In other works, we are looking for a solution to the following equation in the unknown process (◊1 , ◊2 , · · · , ◊n ), µit ≠ rt = n ÿ i = 1, 2, 3, · · · , m, t œ [0, T ]. ‡ijt ◊jt , j=1 (36) We call these equations market price of risk equations. We now consider the portfolio process X, which comprises = ( 1 , 2 , · · · , n ) shares of stocks and amount of money. The dynamics of the portfolio process can be given by dXt = m ÿ it dSit i=1 = rt Xt dt + + rt (Xt ≠ m ÿ i=1 it Dt m ÿ it Sit )dt i=1 d(Dt Sit ), which implies dDt Xt = m ÿ it d(Dt Sit ), (37) i=1 Lemma 4.10. Let Q be a risk neutral measure, and let X be the value of a portfolio. Under Q, the discounted portfolio value DX is a martingale. Proof. The result follows from the fact that DS is a martingale if Q exists. Definition 4.11. An arbitrage is a portfolio value process X satisfying X0 = 0 and also satisfying for some time T > 0 P(XT Ø 0) = 1, P(XT > 0) > 0. 24 Theorem 4.12 (First fundamental theorem of asset pricing). If a market model has a risk neutral measure, then it does not admit arbitrage. Proof. Consider a portfolio value process X, such that for some time T > 0 P(XT Ø 0), P(XT > 0) > 0. Suppose that there exists a risk neutral measure Q, then we have that X0 = EQ [DT XT ] > 0. This completes the proof. 4.6.3 Uniqueness of the risk-neutral measure In this subsection, we consider the hedging strategies under the assumption that there exists at least one risk neutral measure Q. That is, the systems of equations (36) admits at least one solution ◊ = (◊1 , ◊2 , · · · , ◊n ). It follows from the multidimensional Girsanov theorem (4.7) that we can construct the risk neutral measure Q with this ◊. Under the risk neutral measure Q, {{WjtQ := ⁄ t 0 ◊js ds + Wjt }tœ[0,T ] }j=1,2,··· ,n is a martingale. In the previous subsection, we obtained the result that the market is arbitrage free if and only if there exists a risk neutral measure, though we only proved one side. The existence of a risk neutral measure makes the risk-neutral pricing possible. Before applying the risk neural pricing theory, we need to verify that if the hedging strategy, which backs up the corresponding risk neutral price, exists. Mathematically, we define a security terminated at terminal T by a FT ≠ random variable VT . If we define the risk neutral price of VT by the expectation of discounted payoff under Q, that is, Vt = 1 Q E [DT VT |Ft ], Dt then a natural question here is that, if there is a hedging strategy supporting the price Vt ? To answer this question, we first consider Dt Vt , which is a martingale under Q. Then it follows from the multidimensional martingale representation 4.8 that Dt Vt = V0 + or dDt Vt = ⁄ tÿ n 0 j=1 n ÿ Q js dWjs , Q jt dWjt , j=1 25 t œ [0, T ] t œ [0, T ]. (38) We would like to find a portfolio value process X, which can be used to replicate V . In the other words, we are looking for a X, such that Xt = Vt , t œ [0, T ]. Consider the stochastic differential of Dt Xt , which is given by (37), then we have that dDt Xt = = = m ÿ i=1 m ÿ it d(Dt Sit ) it Dt Sit i=1 n ÿ m ÿ n ÿ ‡ijt (◊jt + dWjt ) j=1 Q it Dt Sit ‡ijt dWjt . (39) j=1 i=1 We let V0 = X0 , then it follows from (38) and (39) that seeking for a hedging strategy = ( 1 , 2 , · · · , m ) (and hence portfolio value process X) is equivalent to looking for the solution to the following system of equations jt Dt = m ÿ i=1 it Sit ‡ijt , j = 1, 2, · · · , n, t œ [0, T ]. The solvability of the above system of equations actually is equivalent to the uniqueness of the solutions to the market price of risk equations (36) or the uniqueness of the risk neutral measure. This equivalence is called second fundamental asset pricing theorem, which is summarised as follows. Definition 4.13. A market model is complete if every derivative security can be hedged. Theorem 4.14 (Second fundamental theorem of asset pricing). Consider a market model that has a risk-neutral probability measure. The model is complete if and only if the risk-neutral probability measure is unique. 26 5 5.1 Change of numeraire Market model In this section, we will work with the multidimensional market model, where the uncertainty is given by n≠ dimensional Brownian motion. In particular, we let Wt = (W1t , W2t , · · · , Wnt ) be a n≠ dimensional Brownian motion defined on a probability space ( , F, {Ft }tØ0 , P), and {Ft }tØ0 the filtration generated by W. There is an adapted interest rate process rt , 0 Æ t Æ T. This can be used to create a money market account whose price per share at time t is st Mt = e And the discount process is given by Dt = e≠ 0 rs ds st 0 . rs ds . There are m primary assets in the market and their prices are given by n ÿ dSit = µit dt + ‡ijt dWjt , Sit j=1 i = 1, 2, · · · , m. (40) where µi and ‡ij are adapted processes. We suppose that there is a unique risk-neutral measure Q, that is, there is a unique n≠dimensional adapted process ◊ = (◊1 , ◊2 , · · · , ◊n ) satisfying the market price of risk equations (36). The risk neutral measure is constructed using the multidimensional Girsanove Theorem (4.7). Under Q, the Bronian motion WjtQ = Wjt + ⁄ t 0 ◊js ds, j = 1, 2, · · · , n, are independent of one another. Moreover, the uniqueness of the risk neutral measure, according to the second fundamental asset pricing theorem, implies the completeness of the market model, which means that every security can be hedged. Thus, the discounted process of any security in the complete market is a martingale under the risk neutral measure. 5.2 Numeraire A numeraire is the unit of account in which other assets are denominated. For example, We say the measure Q is risk-neutral for the money market account numeraire. We shall see that sometimes it is convenient to change the numeraire because of modeling considerations as well. A model can be complicated or simple, depending on the choice of the numeraire for the model. In principle, we can take any positively priced asset as a numeraire and denominate all other assets in terms of the chosen numeraire. The asset we take as 27 numeraire could be one of the primary assets given by (40) or it could be a derivative asset. Regardless of which asset we take, it has the stochastic representation provided by the following theorem. Theorem 5.1 (Stochastic representation of assets). Let N be a strictly positive price process for a non-dividend-paying asset, either primary or derivative, in the multidimensional market model. Then there exists a vector volatility process ‹ = (‹1 , ‹2 , · · · , ‹n ) such that dNt = rt Nt dt + Nt n ÿ ‹jt dWjtQ . (41) j=1 The equation is equivalent to each of the equations dDt Nt = Dt Nt n ÿ ‹jt dWjtQ , (42) j=1 Dt Nt = N0 exp{ ⁄ tÿ n 0 j=1 ⁄ tÿ n Nt = N0 exp{ where ||‹t || = qn j=1 0 j=1 1⁄ t ||‹s ||2 ds}, 2 0 (43) 1 (rs ≠ ||‹s ||2 )ds}, 2 (44) Q ‹js dWjs ≠ Q ‹js dWjs + ⁄ t 0 |‹jt |2 , ’t œ [0, T ]. Proof. Under the risk neutral measure Q, DN is a martingale. Therefore, the martingale representation theorem yields that, there exists an adapted process = ( 1 , 2 , · · · , n ), such that dDt Nt = n ÿ Q jt dWjt . j=1 Since Nt > 0, ’t œ [0, T ], we let ‹jt = see that Dt Nt follows (42). Note that Nt = Mt (Dt Nt ), then jt Dt Nt , ’j = 1, 2, · · · , n, t œ [0, T ]. Then it is to dNt = rt Mt Dt Nt dt + Mt dDt Nt , (41) follows from (42) and the fact Mt Dt = 1. Finally, (43) and (44) follows from the application of Ito’s formula on (42) and (41) respectively. 28 We can use multidimensional Girsanov theorem to change the measure. Define WjtN = ≠ ⁄ t 0 ‹js ds + WjtQ , j = 1, 2, · · · , n, and a new probability measure 1 ⁄ P (A) = DT NT dQ, ’A œ F. N0 A N Then it follows from the multidimensional Girsanov theorem that W N = (W1N , W2N , · · · , WnN ) is a Brownian motion under measure PN . Moreover, for an arbitrary random variable X, the expectation of X under PN is given by EN [X] = 1 Q E [DT NT X]. N0 More generally, Dt Nt DT NT = EQ [ |Ft ], N0 N0 0 Æ t Æ T, is the Radon-Nikodym derivative process, which is similar to Zt in the Girsanov theorem. Therefore, for a random variable Y , which is Ft ≠ measurable and integrable, we have that EN [Y |Fs ] = 1 EQ [Y Dt Nt |Fs ], Ds Ns 0 Æ s Æ t Æ T. Theorem 5.2. Let St and Nt be the prices of two assets and let ‡ = (‡1 , ‡2 , · · · , ‡n ) and ‹ = (‹1 , ‹2 , · · · , ‹n ) denote their respective volatility vector processes: dDt St = Dt St n ÿ ‡jt dWjtQ , j=1 dDt Nt = Dt Nt n ÿ ‹jt dWjtQ , j=1 where ‡ and nu are adapted processes. Take Nt as the numeraire, then the price of St becomes StN = measure PN , the process S N is a martingale. Moreover, dStN = StN n ÿ j=1 (‡jt ≠ ‹jt )dWjtN . 29 St . Nt Under the Proof. We have that ⁄ tÿ n Dt St = S0 exp{ Dt Nt = N0 exp{ 0 j=1 Q ‡js dWjs ≠ ⁄ tÿ n 0 j=1 1⁄ t ||‡s ||2 ds}, 2 0 Q ‹js dWjs ≠ 1⁄ t ||‹s ||2 ds}. 2 0 and hence StN ⁄ tÿ n S0 1⁄ t Q = exp{ (‡js ≠ ‹js )dWjs ≠ (||‡s ||2 ≠ ||‹s ||2 )ds}. N0 2 0 0 j=1 Define Xt = ⁄ tÿ n 0 j=1 Q (‡js ≠ ‹js )dWjs ≠ 1⁄ t (||‡s ||2 ≠ ||‹s ||2 )ds. 2 0 Then it is to see that dXt = 1 (‡jt ≠ ‹jt )dWjtQ ≠ (||‡t ||2 ≠ ||‹t ||2 )dt, 2 j=1 n ÿ and dXt dXt = n ÿ j=1 With f (x) = S0 x e , N0 (‡jt ≠ ‹jt )2 dt. we have that StN = f (Xt ), then dStN = StN ( n ÿ j=1 (‡jt ≠ ‹jt )(dWjtQ ≠ ‹jt )) = StN ( n ÿ j=1 (‡jt ≠ ‹jt )dWjtN ). This completes the proof. 5.3 Forward measure Although there may be multiple Brownian motions driving the model of this section, in order to simplify the notation, we assume in this section that there is only one. It is not difficult to rederive the results presented here under the assumption that there are n≠Brownian motions. 1311 Blt 不⼆六 a strictly positive Consider a zero coupon bond terminated at T . The bond isEQIDHFD asset, and the discounted price Dt B(t, T ) given by (31) is a martingale under the risk neutral measure Q. It follows from Theorem 5.2 that there exists an adapted process ‡t (T ) in t such that d(Dt B(t, T )) = Dt B(t, T )‡t (T )dWtQ . 30 Definition 5.3. Let T be a maturity date. We define the T ≠forward measure PT by PT (A) = ⁄ A DT dQ, B(0, T ) ’A œ F. It follows from the Girsanov theorem that WtT = WtQ ≠ ⁄ t 0 ‡t (T )dWtQ is a Brownian motion under PT . For a contingent claim VT , which is FT ≠measurable, we have that 1 EQ [VT DT |Ft ] Dt B(t, T ) 1 = Vt , B(t, T ) ET [VT |Ft ] = which yields that Vt = B(t, T )ET [VT |Ft ]. (45) Equation (45) illustrates that if we can find a simple dynamics of the underlying asset under the T ≠forward measure, then we only need to consider the estimate of VT , and hence we do not have to consider the correlation between the discount factor DT and VT . 5.4 The Black-Scholes formula with random interest rates We present a generalized Black-Scholes option pricing formula that permits the interest rate to be random. The classical Black-Scholes assumption that the volatility of the underlying asset is constant is here replaced by the assumption that the volatility of the forward price of the underlying asset is constant. Because the forward price is a martingale under the forward measure, and W T (t) is the Brownian motion used to drive asset prices under the forward measure, the assumption of constant volatility for the forward price is equivalent to the assumption dF orS (t, T ) = ‡F orS (t, T )dWtT , (46) where ‡ is a constant. The bond maturity T is chosen to coincide with the expiration time T of the option. Theorem 5.4. Let S be the price of an asset denominated in currency, and assume the forward price of this asset satisfies (46) with a positive constant ‡. The value at 31 time t œ [0, T ] of a European call on this asset, expiring at time T with strike price K, is Vt = St N (d1 ) ≠ KB(t, T )N (d2 ), where 1 F orS (t, T ) 1 2 d1,2 = Ô (ln ± ‡ (T ≠ t)). K 2 ‡ T ≠t Proof. It follows from the risk neutral pricing theory that Vt = 1 Q E [DT (ST ≠ K)+ |Ft ]. Dt Then (45) yields that Vt = B(t, T )ET [(ST ≠ K)+ |Ft ]. Note that ST = F orS (T, T ), then the representation above becomes Vt = B(t, T )ET [(F orS (T, T ) ≠ K)+ |Ft ]. Since the dynamics of FS (t, T ) follows (46), then the standard Black-Scholes formula (22) yields that ET [(F orS (T, T ) ≠ K)+ |Ft ] = F orS (t, T )N (d1 ) ≠ KN (d2 ) Moreover, it follows from (32) that St = F orS (t, T )B(t, T ). Then Vt = B(t, T )(F orS (t, T )N (d1 ) ≠ KN (d2 )) = St N (d1 ) ≠ KB(t, T )N (d2 ). 32 6 6.1 Interest rate models Short rate models Consider a probability space ( , F, P). Suppose that W is a one dimensional Brownian motion and the filtration generated by the Brownian motion is {Ft }tØ0 . We assume the completeness of the market model, i.e., there exists a unique risk measure Q, such that for every asset, either primary or derivative, the discounted price is a martingale under Q. Suppose that the interest rate rt is governed by the following stochastic differential equation drt = b(t, rt )dt + ‡(t, rt )dWtQ , (47) where W Q is a Brownian motion under Q and b, ‡ are continuous deterministic functions such that SDE (47) admits a unique strong solution. 6.1.1 Vasicek model In the Vasicek model (Vasicek 1977), the interest rate rt follows the differential equation given by drt = (b + —rt )dt + ‡dWtQ , (48) where b, —, ‡ are constant and — is negative. We consider the zero coupon bond price under Vasicek model. It follows from the risk neutral pricing theory that the zero coupon bond price P (t, r; T ) can be given by P (t, r; T ) = EQ [e≠ sT t rs ds |rt = r] where t, r, T mark the current time, current value of the interest rate and the maturity date of the bond. It follows from the Fyenman-Kac formula that the price function P (t, r; T ) solves the partial differential equation problem as follows. ˆP 1 ˆ2P ˆP + ‡ 2 2 + (b + —r) ≠ rP = 0 r œ (≠Œ, Œ), ˆt 2 ˆr ˆr with the terminal condition P (T, r; T ) = 1. Conjecture that P (t, r; T ) = exp(≠A(t) ≠ B(t)r). 33 (49) Plug the representation of the bond price P into the PDE, then we have dA(t) ‡2 = B(t)2 ≠ bB(t), dt 2 dB(t) = ≠—B(t) ≠ 1, dt t œ [0, T ), t œ [0, T ). In addition, the terminal condition of the PDE problem yields the the terminal conditions of the ODEs above, i.e., A(T ) = 0, B(T ) = 0. Therefore, solving the above ODE system, we have that B(t) = A(t) = 1 —(T ≠t) (e ≠ 1), — ‡ 2 (4e—(T ≠t) ≠ e2—(T ≠t) ≠ 2—(T ≠ t) ≠ 3) e—(T ≠t) ≠ 1 ≠ —(T ≠ t) + b . 4— 3 —2 Substitute the representation of A(t) and B(t) into (49), then we obtain the closed form formula for a zero coupon bond under Vasicek model. In addition, it follows from the dynamics of the Vasicek model (48) and Ito’s formula that b EQ [rt |r0 = x] = xe—t + (e—t ≠ 1), — V arQ (rt |r0 = x) = 6.1.2 ‡ 2 2—t (e ≠ 1) 2— Cox-Ingersoll-Ross model (CIR) modelú In the CIR model (Cox et al. 2005), the dynamics of the interest rate rt is given by Ô drt = Ÿ(◊ ≠ rt )dt + ‡ rt dWtQ . (50) where the constant parameters Ÿ Ø 0, ◊ Ø 0, ‡ > 0. In contrast to the Vasicek model in which the interest rate will take values in (≠Œ, Œ), the CIR model can keep the interest rate process always above 0, which makes the model more realistic. In particular, if 2Ÿ◊ > ‡ 2 , then rt will never reach 0, for any r0 > 0. Otherwise, rt will occasionally touch 0. 34 It follows from risk neutral pricing theory that the bond price P (t, r; T ) can be given by P (t, r; T ) = EQ [e≠ Then the Feynman-Kac formula yields that sT t rs ds |rt = r]. ˆP 1 ˆ2P ˆP + ‡ 2 r 2 + Ÿ(◊ ≠ r) ≠ rP = 0 r œ (≠Œ, Œ), ˆt 2 ˆr ˆr with the terminal condition P (T, r; T ) = 1. Conjecture that P (t, r; T ) = exp(≠A(t) ≠ B(t)r). (51) Plug the representation of the bond price P into the PDE, then we have dA(t) = ≠Ÿ◊B(t), dt t œ [0, T ), dB(t) ‡2 = B(t)2 + ŸB(t) ≠ 1, t œ [0, T ). dt 2 In addition, the terminal condition of the PDE problem yields the the terminal conditions of the ODEs above, i.e., A(T ) = 0, B(T ) = 0. Therefore, solving the above (Riccati) ODE system, we have that B(t) = 2(e“(T ≠t) ≠ 1) , (“ + Ÿ)e“(T ≠t) + “ ≠ Ÿ (“+Ÿ)(T ≠t) 2 ≠2Ÿ◊ 2“e A(t) = log( ), 2 “(T ≠t) ‡ (“ + Ÿ)(e ≠ 1) + 2“ Ô where “ = Ÿ2 + 2‡ 2 . Substitute the representation of A(t) and B(t) into (51), then we obtain the closed form formula for a zero coupon bond under CIR model. In addition, it follows from the dynamics of the CIR model (50) and Ito’s formula that EQ [rt |r0 = x] = xe≠Ÿt + ◊(1 ≠ e≠Ÿt ), V arQ (rt |r0 = x) = x ‡2◊ ‡ 2 ≠Ÿt (e ≠ e≠2Ÿt ) + (1 ≠ e≠Ÿt )2 . Ÿ 2Ÿ 35 6.2 Heath-Jarrow-Morton (HJM) model Short rate models are not always flexible enough to calibrating them to the observed initial term-structure. In the late eighties, Heath, Jarrow and Morton (Heath et al. 1992) proposed a new framework for modelling the entire forward curve directly. This section provides the essentials of the HJM framework. 6.2.1 Forward rates Let us fix a time horizon T̄ . All bonds in the following discussing will mature at or before time T̄ . We assume that all bonds bear no risk of default. We denote that price at time t of a zero coupon bond maturing at time T Æ T by B(t, T ). We assume further that for every t and T such that 0 Æ t Æ T Æ T̄ , the bond price B(t, T ) is defined. In addition, we suppose that the interest rate is strictly positive between times t and T , then B(t, T ) is strictly less than one whenever t < T. Definition 6.1. We define the forward rate at time t for investing at time T to be f (t, T ) = ≠ ˆ log B(t, T ) . ˆT intuitively, f (t, T ) is the instantaneous interest rate at time T that can be locked in at the earlier time t. Moreover, it is easy to see that B(t, T ) = e≠ sT t f (t,v)dv , which implies that it does not appear to matter (at least theoretically) whether we build a model for forward rates or for bond prices. In addition, the interest rate at time t is rt = f (t, t). 6.2.2 Dynamics of forward rates and bond prices Assume that f (0, T ) is known at time 0. We call this the initial forward rate curve. In HJM model, the forward rate at later times t for investing at still later times T is given by f (t, T ) = f (0, T ) + or ⁄ t 0 –(u, T )du + ⁄ t 0 ‡(u, T )dWu , df (t, T ) = –(t, T )dt + ‡(t, T )dWt , (52) where W is a Brownian motion under the actual measure P and –, ‡ are adapted stochastic processes in t, representing the drift and volatility of f under the actual measure P. 36 Consider the dynamics of ≠ d(≠ ⁄ t 0 st 0 f (t, v)dv. It follows from simple calculation that f (t, v)dv) = f (t, t)dt ≠ ⁄ T = rt ≠ t ⁄ T t df (t, v)dv (–(t, v)dt + ‡(t, v)dWt )dv We reverse the order of the integration, writing ⁄ T t ⁄ T –(t, v)dtdv = ‡(t, v)dWt dv = t ⁄ T t ⁄ T t –(t, v)dvdt = –ú (t, T )dt, ‡(t, v)dvdWt = ‡ ú (t, T )dWt , where –ú (t, T ) = ⁄ T –(t, v)dv, ‡ ú (t, T ) = ⁄ T ‡(t, v)dv. t t In conclusion, we have that d(≠ ⁄ T t Note that B(t, T ) = e ≠ f (t, v)dv) = rt dt ≠ –ú (t, T )dt ≠ ‡ ú (t, T )dWt . sT t f (t,v)dv . Then Ito’s formula yields that 1 dB(t, T ) = B(t, T )(rt ≠ –ú (t, T ) + ‡ ú (t, T )2 )dt ≠ ‡ ú (t, T )B(t, T )dWt . 2 6.2.3 No arbitrage condition The HJM model has a zero coupon bond with maturity T , for every T Æ T̄ . We need to make sure there is no arbitrage opportunity in the market. According the Girsanov theorem that it is equivalent to guarantee that the risk neutral measure Q exists. Consider the dynamics of Dt B(t, T ). It follows from Ito’s formula that 1 dDt B(t, T ) = Dt B(t, T )((≠–ú (t, T ) + ‡ ú (t, T )2 )dt ≠ ‡ ú (t, T )dWt ) 2 (53) We want to write the above equation in the form of dDt B(t, T ) = ≠‡ ú (t, T )Dt B(t, T )(◊t + dWt ). 37 (54) In other words, we will find a process ◊, such that WtQ = ⁄ t 0 ◊s ds + Wt , (55) is a Brownian motion under the measure Q, which is defined by Q(A) = ⁄ A 1 e≠ 2 s T̄ 0 ◊s2 ds≠ s T̄ 0 ◊s dWs , ’A œ FT̄ . Comparing (53) and (54), we must solve the equation 1 ≠‡ ú (t, T )◊t = ≠–ú (t, T ) + ‡ ú (t, T )2 2 (56) Recall the definitions of –ú (t, T ) and ‡ ú (t, T ) and differentiate the above equation with respect to T , we then have –(t, T ) = ‡(t, T )(‡ ú (t, T ) + ◊t ). (57) Here ◊t is called market price of risk. Theorem 6.2 (HJM no arbitrage condition). With the above notations, a termstructure model for zero coupon bond prices of all maturities in [0, T̄ ] and driven by a one dimensional Brownian motion does not admit arbitrage if there exists a process ◊, such that (57) holds for all 0 Æ t Æ T Æ T̄ . Proof. It remains to check that if ◊t solves (56), then we can use Girsanov theorem to construct the risk neutral measure. The existence of the risk neutral measure guarantees the absence of arbitrage. Suppose that ◊t solves (57), then integrating with respect to the second variable from t to T , we have that 1 1 –ú (t, T ) ≠ –ú (t, t) = (‡ ú (t, T ))2 ≠ (‡ ú (t, t))2 + ‡ ú (t, T )◊t ≠ ‡ ú (t, t)◊t . 2 2 Note that –ú (t, t) = ‡ ú (t, t) = 0, we have that 1 –ú (t, T ) = ‡ ú (t, T )2 + ‡ ú (t, T )◊t , 2 which completes the proof. As long as ‡(t, T ) is non-zero, we can solve (57) for ◊t : ◊t = –(t, T ) ≠ ‡ ú (t, T ), ‡(t, T ) 0 Æ t Æ T. This shows that ◊t is unique and hence the risk neutral measure is unique. In this case, it follows from the second fundamental theorem of asset pricing that the market model is complete. 38 6.2.4 HJM under risk neutral measure In this subsection, we consider the dynamics of f (t, T ) and B(t, T ). For the forward rate f (t, T ), whose dynamics is defined by (52), we have that df (t, T ) = –(t, T )dt + ‡(t, T )dWt = ‡(t, T )‡ ú (t, T )dt + ‡(t, T )(◊t + dWt ) = ‡(t, T )‡ ú (t, T )dt + ‡(t, T )dWtQ where W Q is defined by (55). Turning to the bond price B(t, T ), we know that dDt B(t, T ) = ≠‡ ú (t, T )Dt B(t, T )(◊t + dWt ) = ≠‡ ú (t, T )Dt B(t, T )dWtQ . Note that B(t, T ) = 1 (Dt B(t, T )), Dt then it follows from Ito’s formula that dB(t, T ) = rt B(t, T )dt ≠ ‡ ú (t, T )B(t, T )dWtQ . The following theorem summarises this discussion. Theorem 6.3 (Term structure evolution under risk neutral measure). Under the HJM no arbitrage condition (57), the foward rates evolve according to the equation df (t, T ) = ‡(t, T )‡ ú (t, T )dt + ‡(t, T )dWtQ , and the zero coupon bond prices evolve according to the equation dB(t, T ) = rt B(t, T )dt ≠ ‡ ú (t, T )B(t, T )dWtQ , or equivalently, B(t, T ) = B(0, T ) exp{ ⁄ t 0 rs ds ≠ ⁄ t 0 ‡ ú (s, T )dWsQ 1⁄ t ú ≠ (‡ (s, T ))2 ds}. 2 0 (58) where W Q is a Brownian motion under the risk neutral measure Q. 6.3 Forward LIBOR models In this section, we derive the Black formula for the interest rate cap, a common fixed income derivative underlain on the LIBOR (London Interbank Offered Rate). The idea is similar to the one introduced in Section 5.4, that is, We simplifies the structure of the pay-off function by introducing a forward measure, so that the pricing formula of the fixed income derivative can be obtained within the standard Black-Scholes framework. 39 6.3.1 Motivation Consider the HJM model that satisfies the no arbitrage condition, df (t, T ) = ‡(t, T )‡ ú (t, T )dt + ‡(t, T )dWtQ , where W Q is a Brownian motion under the risk neutral measure Q and ‡(t, T ), ‡ ú (t, T ) are defined in the previous section. We want to choose a ‡(t, T ) such that f (t, T ) Ø 0. A natural candidate is a geometric Brownian motion, in which ‡(t, T ) = ‡fs(t, T ), where the constant ‡ > 0. In this case, the drift ‡(t, T )‡ ú (t, T ) = ‡ 2 f (t, T ) tT f (t, v)dv and the dynamics of the forward interest rate f becomes df (t, T ) = ‡ 2 f (t, T ) ⁄ T t f (t, v)ddt + ‡f (t, T )dWtQ , It is worth noting that the fast growth of the drift term will cause the explosion of stochastic differential equation. This difficulty with continuously compounding forward rates causes us to introduce forward LIBOR. 6.3.2 LIBOR and forward LIBOR Let 0 Æ t Æ T and ” > 0 be given. Suppose that B(t, T ) is the price of a bond with maturity date T and principal 1 at time t œ [0, T ]. Then at time t, in order to lock in the interest rate during [T, T + ”], one can hold a short position of size 1 in B(t,T ) a T ≠zero coupon bond and a long position of size B(t,T in a T + ” zero coupon +”) bond. This position can be created at zero cost at time t, it calls for ”investment” B(t,T ) of 1 at time T to cover the short position, and it repays B(t,T at time T + ”. +”) Suppose that money will be accumulated according to the simple interest rate rule, then during the time interval t, T + ”, the simple interest rate is given by L(t, T ) = B(t, T ) ≠ B(t, T + ”) ”B(t, T + ”) (59) Definition 6.4. We call L(t, T ) defined by (59) forward LIBOR, ’0 Æ t < T . Particularly, when t = T , L(T, T ) is LIBOR set at time T . ” is called the tenor of the LIBOR. 6.3.3 Pricing a backset LIBOR contract An interest rate swap is an agreement between two parties A and B that A will make fixed interest rate payments on some “notional amount” to B at regularly spaced dates and B will make variable interest rate payments on the same notional amount on these same dates. The variable rate is often backset LIBOR, defined on one payment date to be the LIBOR set on the previous payment date. The no-arbitrage price of a payment of backset LIBOR on a notional amount of 1 is given by the following theorem. 40 Theorem 6.5 (Price of backset LIBOR). Let ” > 0, 0 Æ t Æ T be given. The no arbitrage price at time t of a contract that pays L(T, T ) at time T + ” is given by S(t) = B(t, T + ”)L(t, T ). Proof. It follows from the risk neutral pricing theory that S(t) = EQ [e≠ s T +” t rs ds L(T, T )|Ft ]. Then the definition of LIBOR (59) yields that ≠ B(T, T + ”) |Ft ] ”B(T, T + ”) s T +” s T +” 1 1 = EQ [e≠ t rs ds |Ft ] ≠ EQ [e≠ t rs ds |Ft ] ”B(T, T + ”) ” sT s T +” s T +” 1 1 = EQ [e≠ t rs ds EQ [e≠ T rs ds |FT ] |Ft ] ≠ EQ [e≠ t rs ds |Ft ]. ”B(T, T + ”) ” (60) S(t) = E [e Q ≠ s T +” t rs ds 1 As B(T, T + ”) = EQ [e≠ yields that s T +” T rs ds S(t) = |FT ], B(t, T + ”) = EQ [e≠ B(t, T ) ≠ B(t, T + ”) . ” s T +” t rs ds |Ft ], then (60) (61) Combining (61) with the definition of L(t, T ), we obtain S(t) = L(t, T )B(t, T + ”). This concludes the proof. 6.3.4 Black caplet formula A common fixed income derivative security is an interest rate cap, a contract that pays the difference between a variable interest rate applied to a principal and a fixed interest rate (a cap) applied to the same principal whenever the variable interest rate exceeds the fixed rate. let the tenor ”, the principal (also called the notional amount) P, and the cap K be fixed positive numbers. An interest rate cap pays (”P L(”j , ”j ) ≠ K)+ at time ”(j + 1), for j = 0, 1, · · · , n. To determine the price at time zero of the cap, it suffices to price one of the payments, a so-called interest rate caplet, and then sum these prices over the payments. We show here how to do this and obtain the Black caplet formula. We also note that each of these payments is K of the form ”P (L(”j , ”j ) ≠ K Õ )+ , where K Õ = ”P . Thus, it suffices to determine the + time zero price of the payment (L(T, T ) ≠ K) at time T + ” for an arbitrary T and K > 0. 41 It follows from Theorem 6.5 that L(t, T ) = us to introduce the forward measure P T +” S(t) . B(t,T +”) This observation motivates ⁄ 1 (A) = DT +” dQ, ’A œ FT +” . B(0, T + ”) A under which WtT +” = ⁄ t 0 ‡(s, T + ”)ds + WtQ (62) is a Brownian motion. S(t) Theorem 5.2 implies that B(t,T is a martingale under PT +” . According to the +”) martingale representation, there exits an adapted process “(t, T ), a process in t, such that dL(t, T ) = “(t, T )L(t, T )dWtT +” . (63) Theorem 6.6 (Black caplet formula). Consider a caplet that pays (L(T, T ) ≠ K)+ at T + ”, where K is a non-negative number. Suppose “ in (63) is non-random, then the price of the caplet at time zero is given by B(0, T + ”)(L(0, T )N (d1 ) ≠ KN (d2 )) where d1,2 L(0, T ) 1 ⁄ T = sT Ò (log ± “(t, T )2 dt). 2 K 2 0 “(t, T ) dt 1 0 Proof. It follows from the risk neutral pricing theorem that the price of the caplet is given by DT +” (L(T, T ) ≠ K)+ ] B(0, T + ”) = B(0, T + ”)ET +” [(L(T, T ) ≠ K)+ ] EQ [DT +” (L(T, T ) ≠ K)+ ] = B(0, T + ”)EQ [ Then the Conclusion follows from the dynamics of L(t, T ) (63) and the standard Black-Scholes formula. 6.3.5 Forward LIBOR and zero coupon bond volatilities In the end of this section, we derive the process “(t, T ). According to the definition of the fowrd LIBOR, L(t, T ) can be rewritten as L(t, T ) = B(t, T ) 1 ≠ . ”B(t, T + ”) ” 42 It follows from (58) that ⁄ t B(t, T ) = exp{≠ (‡ ú (s, T ) ≠ ‡ ú (s, T + ”))dWsQ B(t, T + ”) 0 ⁄ t 1 ≠ ((‡ ú (s, T ))2 ≠ (‡ ú (s, T + ”))2 )ds} 2 0 Then Ito’s formula yields that 1 dL(t, T ) = (L(t, T ) + )(‡ ú (t, T + ”) ≠ ‡ ú (t, T ))(‡ ú (t, T + ”)dt + dWtQ ). ” Together with (62), we have that 1 dL(t, T ) = (L(t, T ) + )(‡ ú (t, T + ”) ≠ ‡ ú (t, T ))dWtT +” . ” Comparing with (63), we obtain “(t, T ) = 1 + ”L(t, T ) ú (‡ (t, T + ”) ≠ ‡ ú (t, T )). ”L(t, T ) 43 7 7.1 Foreign exchange models The market model The model is driven by a two dimensional Brownian motion (W1 , W2 ) defined on the probability space ( , F, P), where P is a probability measure in the actual world. We suppose that there are two currencies in the market. Particularly, the stock price in the domestic currency follows the stochastic differential equation dSt = µt St dt + ‡t St dW1t . The domestic money account is accumulated with the continuously compounding interest rate rt , hence the domestic money account and the domestic discount factor can be respectively given by st Mt = e 0 rs ds , Dt = e ≠ st rs ds 0 . We assume that the foreign interest rate is rtf , which leads to the foreign money account and foreign discount factor are respectively given by Mtf st =e 0 rsf ds , Dtf =e ≠ st 0 rsf ds . Finally, we suppose that foreign exchange rate Q, which gives the units of domestic currency per unit of foreign currency, satisfies dQt = “t Qt dt + ‡2t Qt (flt dW1t + Ò 1 ≠ fl2t dW2t ) If we measure the foreign account by the domestic currency, then the value of the foreign money account is given by M̄tf := Qt Mtf . Assume that fl, ‡1 , ‡2 “, µ, r, rf are adapted processes with respect to the filtration {Ft }tØ0 generated by the Brownian motion (W1 , W2 ). Moreover, we suppose that ‡1 > 0, ‡2 > 0, ≠1 < fl < 1. Define W3t = ⁄ t 0 fls dW1s + ⁄ tÒ 0 1 ≠ fl2s dW2s Then it follows Levy’s characterisation that W3 is a Brownian motion under P. In addition, by simple calculation, we have that dW1t dW3t = flt dt, and hence dSt dQt = flt ‡1t ‡2t dt. St Qt 44 7.1.1 Domestic risk neutral measure There are three assets that can be traded: the domestic money market account, the stock, and the foreign money market account. We shall price each of these in domestic currency and discount at the domestic interest rate. The result is the price of each of them in units of the domestic money market account. Under the domestic risk-neutral measure, all three assets priced in units of the domestic money market account must be martingales. We use this observation to find the domestic risk-neutral measure. We note that the first asset, the domestic money market account, when priced in units of the domestic money market, has constant price 1. This is always a martingale, regardless of the measure being used. We consider the stock price St . It is to see that the stock price satisfies that dSt = rt St dt + ‡1t St (◊1t dt + dW1t ), where ◊1t is the solution to the following market price of risk equation ‡1t ◊1t = µt ≠ rt , which implies, µt ≠ r t . ‡1t ◊1t = To obtain the domestic risk neutral measure Q, we construct W1Q , which will be the Brownian motion under Q, as follows. W1tQ = ⁄ t 0 ◊1s ds + W1t . (64) The third asset that can be traded is foreign currency money account, whose dynamics is given by dM̄tf = dMtf Qt = (“t + rtf )Mtf Qt dt + Mtf Qt ‡2t (flt dW1t Substitute (64) into the equation above, we then obtain dM̄tf = rt M̄tf dt + (“t + rtf ≠ rt ≠ ‡2t flt ◊1t )M̄tf dt + + Ò 1 ≠ fl2t dW2t ). M̄tf ‡2t (flt dW1tQ + Ò 1 ≠ fl2t dW2t ) (65) To construct the risk neutral measure, we would like to choose ◊2t , such that W2tQ and dM̄tf = rt M̄tf dt + = ⁄ t 0 ◊2s ds + W2t , ‡2t M̄tf (flt dW1tQ 45 + (66) Ò 1 ≠ fl2t dW2tQ ) (67) Combining (65), (66) and (67), we obtain the second market price of risk equation as follows, “t + rtf ≠ rt = ‡2t (flt ◊1t + Solving the equation above, we obtain that ◊2t = Ò 1 ≠ fl2t ◊2t ). “t + rtf ≠ rt ≠ ‡2t flt ◊1t Ò ‡2t 1 ≠ fl2t . We are now using the Girsanov theorem to construct the risk neutral measure. Define Zt = exp{≠ ⁄ t 0 ⁄ t ⁄ t 1 2 2 (◊1s + ◊2s )ds ≠ ◊1s dW1s ≠ ◊2s dW2s }. 2 0 0 Construct the risk neutral measure Q, as follows, Q(A) = ⁄ A ZT dP, A œ FT , Then the Girsanov theorem yields that (W1Q , W2Q ) is a Brownian motion under Q and hence the discounted prices Dt Mtf , Dt St are martingales under Q. In addition, it is easy to verify that W3t = ⁄ t 0 Q fls dW1s + ⁄ tÒ 0 Q 1 ≠ fl2s dW2s is a Brownian motion under Q. Thus, we can write the dynamics of the foreign exchange rate Q in the following form, dQt = (rt ≠ rtf )Qt dt + ‡2t Qt dW3tQ (68) Particularly, we also have that dW3tQ dW1tQ 7.1.2 = flt dt, dW3tQ dW2tQ = Ò 1 ≠ fl2t dt Foreign risk neutral measure Similar to the previous subsection, we here would like to choose a risk neutral measure, such that the discounted domestic money account, discounted stock price and discounted foreign money account are martingales, where the discount factor is given by Dtf and the currency unit is the foreign. First, it is easy to see that the foreign money account, when priced and discounted at the foreign currency, is always 1. Therefore, the discounted foreign money account at the foreign currency is a martingale. 46 We consider the stock price and the domestic money account. We note that at the foreign currency, the stock price and the domestic money account can be given by Dtf St St = f , Qt Mt Qt Dtf Mt Mt = f . Qt Mt Qt Note that Dt Mtf Qt , Dt St and Dt Mt = 1 are martingales under Q. This observation Df S Dtf Mt and Theorem 5.2 imply that Qt t t and Q are martingales under Pf , where t Pf (A) = ⁄ DT M f QT dQ. Q0 A Particularly, the process W f = (W1f , W2f ) given by W1tf =≠ W2tf = ≠ ⁄ t 0 ⁄ t ‡2s fls ds + W1tQ 0 Ò ‡2s 1 ≠ fl2s ds + W2tQ is a two dimensional Brownian motion under Pf . Here Pf is said to be the foreign risk neutral measure. Similarly, we can introduce the Brownian motion (under Pf ) W3f given by W3tf = ⁄ t 0 f fls dW1s + It is easy to verify that W3tf = ≠ ⁄ t 0 ⁄ tÒ 0 f 1 ≠ fl2s dW2s . ‡2s ds + W3tQ , and dW3tf dW1tf = flt dt, dW3tf dW2tf = Finally, it follows from (5.2) or Ito’s formula that Ò 1 ≠ fl2t dt Mt Dtf Mt Dtf d =≠ ‡2t dW3tf , Qt Qt Dtf St Dtf St d = (‡1t dW1tf ≠ ‡2t dW3tf ). Qt Qt 47 7.2 Garman-Kohlhagen formula In this subsection, we assume the domestic and foreign interest rates r and rf and the volatility ‡2 are constant. Consider a call on a unit of foreign currency whose payoff in domestic currency is (QT ≠ K)+ . It follows from the risk neutral pricing theory that at time zero, the value of the option is given by EQ [e≠rT (QT ≠ K)+ ]. In this context, the dynamics of Q given by (68) becomes dQt = (r ≠ rf )Qt dt + ‡2 Qt dW3tQ Using the same techniques in the standard Black-Scholes framework, we obtain that EQ [e≠rT (QT ≠ K)+ ] = e≠r T Q0 N (d1 ) ≠ Ke≠rT N (d2 ), f (69) where d1,2 = 1 Q 1 Ô (log 0 + (r ≠ rf ± ‡22 )T ), K 2 ‡2 T and N (·) is the distribution function of a standard normal distribution. Equation (69) is the Garman-Kohlhagen formula. References Black, F. and M. Scholes (1973): “The Pricing of Options and Corporate Liabilities,” The Journal of Political Economy, 81, 637–654. Cox, J. C., J. E. Ingersoll Jr, and S. A. Ross (2005): “A theory of the term structure of interest rates,” in Theory of valuation, World Scientific, 129–164. Heath, D., R. Jarrow, and A. Morton (1992): “Bond pricing and the term structure of interest rates: A new methodology for contingent claims valuation,” Econometrica: Journal of the Econometric Society, 77–105. Hull, J. (2012): Options, Futures, and Other Derivatives, Pearson, Eighth Editon. Merton, R. C. et al. (1973): “Theory of rational option pricing,” Theory of Valuation, 229–288. Vasicek, O. (1977): “An equilibrium characterization of the term structure,” Journal of financial economics, 5, 177–188. 48