Finite Differences (FD)

As with any class of option, the price of the derivative is governed by solving
the underlying partial differential equation. The use of finite difference
methods allows us to solve these PDEs by means of an iterative procedure.
We can start by looking at the Black-Scholes partial differential equation:

where dV is the change in the value of an option, dt is a small change in time.

is the volatility of the underlying, S is the underlying price and is the
carry (r-q).
By specifying initial and boundary conditions, one can attain numerical
solutions to all the derivatives of the Black-Scholes PDE using a finite
difference grid. The grid is typically set up so that partitions in two
dimensions - space and time (in our case, we would be looking at the asset
price and the change in time):
Once the grid is set up, there are three methods to evaluate the PDE at each
time step. The difference between each of the three methods is contingent
on the choice of difference used for time (i.e. forward, backward or central
differences). Central differences is used for the space grid (S).

Explicit Finite Differences

Explicit Finite Differences

Explicit FD uses forward differences at each time node t. By splitting the differential
equation into the time element and space elements, we can apply forward
differences to time as follows:
if we substitute x = ln(S), the equation becomes:

Applying the finite differences method, the above equation can be broken down and
approximated:
becomes

For the space grid, we can apply central differences for all order of derivatives:
becomes
and
becomes
and
becomes

Explicit Finite Differences

Combining the terms gives:

Which is the same as:

where the probabilities of each of the nodes is:

This case is actually equivalent to the trinomial tree where probabilities can
be assigned to the likelihood of an up move, a down move as well as no
move. It can also be shown that the following approximation holds:

Explicit Finite-Difference
For stability and convergence, x 3t .
As we can see, the result from this method is
explicitly given because we know the value
(the claim) at the boundary where the option
expires. Then, we perform the calculation
backwards in time until the valuation date.
Since the time dependence (i) only depends
on future dates (i+1) we can explicitly
calculate the change, node by node backward
in time.

Implicit Finite Differences

Implicit Finite Differences

The implicit method takes backward differences for the time
derivative but still using central differences for the space derivatives.
Although similar in nature to the explicit finite differences method, the
implicit FD method is typically more stable and convergent than the
explicit FD method - however, it is often more computationally intensive.
The approximation to the PDE under an implicit FD method is given by
the following:

Note that the main difference between the above equation and the one
for the explicit FD method is in the selection of time step i. The
subsequent simplification of the approximation and the associated
probabilities is similar to that of the explicit FD method.

Crank-Nicholson Scheme

Crank-Nicholson Scheme
An improvement over the implicit FD method is the Crank-Nicolson
Scheme which uses central differences for both time and space
dimensions. The result is that over smaller time steps dt, the method
is more accurate, stable and convergent than both implicit and
explicit methods - however, like the implicit FD method (which
requires evaluating equations at each time step) it is more
computationally intensive than the explicit FD method.
The approximation for the PDE is given as: