6-ConceptsProbability Page 192 Wednesday, February 4, 2004 3:00 PM
192 The Mathematics of Financial Modeling and Investment Management
sum of the values taken by the two variables X,Y. Let’s suppose that the
two variables X(ω), Y(ω) have a joint density p(x,y) and marginal densi-
ties pX(x) and pY(x), respectively. Let’s call H the cumulative distribu-
tion of the variable Z. The following relationship holds
Hu()= PZ [ ()ω ≤u]= ∫ ∫ px y ( , )dyxd
A
A = {y ≤–x + u}
In other words, the probability that the sum X + Y be less than or equal
to a real number u is given by the integral of the joint probability distri-
bution function in the region A. The region A can be described as the
region of the x,y plane below the straight line y = –x + u.
If we assume that the two variables are independent, then the distri-
bution of the sum admits a simple representation. In fact, under the
assumption of independence, the joint density is the product of the mar-
ginal densities: p(x,y) = pX(x)pY(x). Therefore, we can write
∞
uy–
Hu()= PZ [ ()ω ≤u]= ∫ ∫px y ( , )dyxd = ∫ ()xd pY y
∫ pX x ()yd
A –∞ –∞
We can now use a property of integrals called the Leibnitz rule,
which allows one to write the following relationship:
∞
dH
-------- = pZ ()u = ∫pX(uy– )pY ()y yd
du –∞
Recall from Chapter 4 that the above formula is a convolution of
the two marginal distributions. This formula can be reiterated for any
number of summands: the density of the sum of n random variables is
the convolution of their densities.
Computing directly the convolution of a number of functions might
be very difficult or impossible. However, if we take the Fourier transforms
of the densities, PZ(s), PX(s), PY(s) computations are substantially simpli-
fied as the transform of the convolution is the product of the transforms:
∞
pZ ()u = ∫ pX(uy– )pY ()y yd ⇒PZ ()s = PX ()s ×PY ()s
- ∞