6-ConceptsProbability Page 197 Wednesday, February 4, 2004 3:00 PM
Concepts of Probability 197
THE REGRESSION FUNCTION
Given a probability space (Ω,ℑ,P), consider a set of p +1 random variables.
Let’s suppose that the random vector {XZ 1 ... Zp} ≡{XZ}, Z = {Z 1 ... Zp}
has the joint multivariate probability density function:
fxz( 1 ...zp)= fx( ,z), z = {z 1 ...zp}
Let’s consider the conditional density
fx( z 1 , ...,zp) = fx( ,z)
and the marginal density of Z,
∞
fz()z = ∫ fx( ,z)dx
- ∞
Recall from an earlier section that the joint multivariate density f(x,z)
factorizes as
fx( ,z) = fx( z)fz ()z
Let’s consider now the conditional expectation of the variable Xgiven Z
= z = {z 1 ... zp}:
∞
g()z = EX[ Z = z] = ∫ vf v( z)dv
- ∞
The function g, that is, the function which gives the conditional expec-
tation of Xgiven the variables Z, is called the regression function. Oth-
erwise stated, the regression function is a real function of real variables
which is the locus of the expectation of the random variable Xgiven
that the variables Z assume the values z.
Linear Regression
In general, the regression function depends on the joint distribution of
[XZ 1 ... Zp]. In financial econometrics it is important to determine
what joint distributions produce a linear regression function. It can be