Conversion of BS PDE to Heat Equation

In its simplest form, the Black-Scholes equation is a PDE of the form

V_t = rV - rx V_x - \frac{1}{2} \sigma^2 x^2 V_{xx} where r, \sigma are positive constants and subscripts correspond to partial differentiation. A solution is a smooth function V = V(t,x) satisfying the BS PDE. Normally, solutions for which x > 0 and 0 \le t \le T are sought. For a particular option pricing problem, solutions must also satisfy the boundary condition V(T,x) = f(x) for some given payoff function f.

It is possible to convert the variable-coefficient PDE into a constant-coefficient PDE by change of variables. Consider the global differeomorphism y = \ln x and let the function W be defined by W(t,y) = V(t,x). Then

V_x = W_y y_x = W_y x^{-1}

V_{xx} = W_{yy} y_x^2 + W_y y_{xx} = W_{yy} x^{-2} + W_y (-x^{-2}).

and so

W_t = rW - r W_y - \frac{1}{2} \sigma^2 (W_{yy} - W_y) = r W - (r - \frac{1}{2} \sigma^2) W_y - \frac{1}{2} \sigma^2 W_{yy}

is a constant-coefficient PDE. Exponential integrating factors can be included to reduce the equation to the heat equation.

Let W(t,y) = U(t,y) e^{at} e^{by}. Then

W_t = (U_t + aU) e^{at} e^{by}

W_y = (U_y + bU) e^{at} e^{by}

W_{yy} = (U_{yy} + 2bU_y + b^2 U)  e^{at} e^{by}

and so

U_t + aU = rU - (r-\frac{1}{2} \sigma^2)(U_y + bU) - \frac{1}{2} \sigma^2 (U_{yy} + 2bU_y + b^2 U)

Choosing

a = r - (r - \frac{1}{2} \sigma^2) b - \frac{1}{2} \sigma^2 b^2 = r + \frac{1}{2} \sigma^2 b^2

0 = -(r - \frac{1}{2} \sigma^2) - \frac{1}{2} \sigma^2 2b

these added degrees of freedom may be used to zero coefficients multiplied by U and U_y to give

U_t = -\frac{1}{2} \sigma^2 U_{yy}

Finally, let \tau = T-t, and S(\tau,y) = U(t,y). Then U_t = - S_\tau and the standard form of the heat equation

S_\tau = \frac{1}{2} \sigma^2 S_{yy}

(where \tau \in [0,T] and y is unconstrained) is obtained. The boundary condition becomes S(y,0) = e^{by} f(e^y).

Finding the Fundamental Solution

[ignore this incorrect development]

For the PDE S_\tau = \frac{1}{2} \sigma^2 S_{yy} and S(y,0) = \delta(y - y_0). Trying S(\tau,y) = A(\tau) B(y),

A'(\tau) B(y) = \frac{1}{2} \sigma^2 A(\tau) B''(y) and so A'(\tau)/ A(\tau) = \frac{1}{2} \sigma^2 B''(y) / B(y), the LHS is a function of \tau and the RHS is a function of y. This can only be simultaneously satisfied when the equation is some constant k. Hence A(\tau) = a e^{k\tau} and B(y) = b e^{k^{1/2} (\sigma^2/2)^{-1/2} y} so S(\tau,y) = ab e^{k\tau + k^{1/2} (\sigma^2/2)^{-1/2} y}. Using the boundary conditon S(0,y_0) = 1 = ab e^{k^{1/2} (\sigma^2/2)^{-1/2} y_0} which requires

[to be continued]

Navigation

About

Raedwulf ….