This preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: Derivative Securities – Fall 2007– Section 7 Notes by Robert V. Kohn, extended and improved by Steve Allen. Courant Institute of Mathematical Sciences. Further discussion of the continuous time framework. Topics in this section: (a) more stochastic calculus; (b) what are the consequences of hedging only at discrete times? (c) the link between riskneutral expectation and PDE’s; and (d) martingales and their importance for option pricing. ******************** More stochastic calculus . You’ll need the following for HW4. Consider a stochastic integral of the form R b a g ( s ) dw ( s ) where g is a deterministic function of s . It has mean value zero – we explained this in Section 6. What about its variance? The answer is simple: E Z b a g ( s ) dw ! 2 = Z b a g 2 ( s ) ds. Here is why. Approximating the stochastic integral by a sum, we see that the square of the stochastic integral is approximately X i g ( s i )[ w ( s i +1 ) w ( s i )] ! X j g ( s j )[ w ( s j +1 ) w ( s j )] = X i,j g ( s i ) g ( s j )[ w ( s i +1 ) w ( s i )][ w ( s j +1 ) w ( s j )] . For i 6 = j the expected value of the i, j th term is 0 since [ w ( s j +1 ) w ( s j )] and [ w ( s i +1 ) w ( s i )] are independent Gaussians, each with mean value 0. For i = j the expected value of the i, j th term is g 2 ( s i )( s i +1 s i ). So the expected value of the squared stochastic integral is approximately X i g 2 ( s i )( s i +1 s i ) , which is a Riemann sum for R b a g 2 ( s ) ds . By the way: since we are assuming that g is deterministic, R b a g ( s ) dw ( s ) is a Gaussian random variable. (Proof: recall that a sum of Gaussians is Gaussian; therefore ∑ i g ( s i )[ w ( s i +1 ) w ( s i )] is Gaussian. Now use the fact that a limit of Gaussians is Gaussian.) Since we know its mean and variance, we have completely characterized this random variable. We are in the habit of focusing on lognormal dynamics, because this is the most basic model for the price of a stock (or the forward price of a stock). Another exactlysolvable SDE is the OrnsteinUhlenbeck process, which solves dy = cydt + σdw, y (0) = y 1 with c and σ constant. (This is not a lognormal process, because the coefficient of dw is not proportional to y .) Ito’s lemma gives d ( e ct y ) = ce ct ydt + e ct dy = e ct σdw so e ct y ( t ) y = σ Z t e cs dw, or in other words y ( t ) = e ct y + σ Z t e c ( s t ) dw ( s ) . We see (using the discussion at the beginning of this section) that y ( t ) is a Gaussian random variable. So it is entirely characterized by its mean and variance. They are easy to compute: the mean is clearly E [ y ( t )] = e ct y since the “dw” integral has expected value 0, and the variance is E h ( y ( t ) E [ y ( t )]) 2 i = σ 2 E " Z t e c ( s t ) dw ( s ) 2 # = σ 2 Z t e 2 c ( s t ) ds = σ 2 1 e 2 ct 2 c ....
View
Full Document
 Fall '11
 Bayou
 Finance, The Land, Mathematical finance, BlackScholes PDE

Click to edit the document details