Multivariate Statistical Distributions
Multivariate Statistical Distributions
Multivariate
Statistical Distributions
1
D.S.G. POLLOCK: ECONOMETRICS
One may wish to characterise the statistical event in terms only of a subset
of the elements in x. In that case, one is interested in the marginal distribution
of the subset.
2
MULTIVARIATE STATISTICAL DISTRIBUTIONS
3
D.S.G. POLLOCK: ECONOMETRICS
Expectations
4
MULTIVARIATE STATISTICAL DISTRIBUTIONS
These are readily established from the definitions (9) and (10). Taken together,
the properties (iii) and (iv) imply that
5
D.S.G. POLLOCK: ECONOMETRICS
The expression for the covariance can be expanded to give C(xi , xj ) = E[xi xj
E(xi )xj E(xj )xi + E(xi )E(xj )] = E(xi xj ) E(xi )E(xj ). By setting xj = xi ,
a similar expression is obtained for the variance V (xi ) = C(xi , xi ). Thus
The property of the expectations operator given under (11)(i) implies that
V (xi ) 0. Also, by applying the property under (11)(v) to the expression for
C(xi , xj ), it can be deduced that
6
MULTIVARIATE STATISTICAL DISTRIBUTIONS
The dispersion matrix is nonnegative definite. This is confirmed via the identity
a0 D(x)a = a0 {E[x E(x)][x E(x)]0 }a = E{[a0 x E(a0 x)]2 } = V (a0 x) 0,
which reflects the fact that variance of any scalar is nonnegative. The following
are some of the properties of the operators:
7
D.S.G. POLLOCK: ECONOMETRICS
N (A1 y; , )kx/yk
(21) = (2)n/2 ||1/2 exp 12 (A1 y )0 1 (A1 y ) kA1 k
= (2)n/2 |AA0 |1/2 exp 12 (y A)0 (AA0 )1 (y A) ;
(23) (z )0 1 (z ) = (y y )0 1 0 1
yy (y y ) + (x x ) xx (x x )
which arises in this particular case. Substituting the RHS into the expression
for N (z; , ) in (19) and using || = |yy ||xx |, gives
N (z; , ) = (2)m/2 |yy |1/2 exp 12 (y y )0 1
yy (y y )
(24) (2)(mn)/2 |xx |1/2 exp 12 (x x )0 1
xx (x x )
= N (y; y , yy ) N (x; x , xx ).
The latter can only be the product of the marginal distributions of y and x,
which proves that these vectors are independently distributed.
The essential feature of the result is that
(26) If z N (, ) is partitioned as
y y yy yx
N , ,
x x xy xx
8
MULTIVARIATE STATISTICAL DISTRIBUTIONS
N (y|x; y + yx 1 1
xx (x x ), yy yx xx xy ).
Proof. The joint distribution of x and y can be factored as the product of the
marginal distribution of x and the conditional distribution of y given x:
wherein
The o-diaogonal blocks of D(w), which are C{y E(y|x), x} = 0 and its
transpose, bear witness to the fact that the prediction error = y E(y|x) is
uncorrelated with x, which is the instrument of the prediction.
The quadratic exponent of the joint distribution of x and y may be ex-
pressed in terms either of z or w. Thus, z 0 1 0 1 0 0 1
zz z = w ww w = z Q ww Qz,
1 01
which indicates that zz = Q ww Q . This is written more explicitly as
yy yx I B0 0 I 0
=
xy xx 0 I 0 xx B I
(32) 0 0
+ B xx B B xx
= .
xx B xx
(33) B = 1
xx xy and = yy yx 1
xx xy .
9
D.S.G. POLLOCK: ECONOMETRICS
which defines the expected value of x for given values of y, is described as the
regression of y on x. The matrix B = 1 xx xy is the matrix of the regression
coecients.
Now that the general the form of the marginal distribution has been estab-
lished, it can be shown that any nondegenerate random vector which represents
a linear function of a normal vector is itself normally distributed. To this end,
it can be proved that
Proof. If B has full row rank, then there exists a nonsingular matrix A0 =
[B 0 , C 0 ] such that
y B
(37) q= = (x b).
z C
10
MULTIVARIATE STATISTICAL DISTRIBUTIONS
The t distribution, which is perhaps the most important of the sampling dis-
tributions, is also extensively tabulated. Again, we shall not give the formula
for the density function; but we may note that the distribution is symmetrical
and that E(t) = 0 and V (t) = n/(n 2). The distribution t(n) approaches the
standard normal N (0, 1) as n tends to infinity. This results from the fact that,
as n tends to infinity, the distribution of the denominator in the ratio defining
the t variate becomes increasingly concentrated around the value of unity, with
the eect that the variate is dominated by its numerator. Finally,
This follows from (30) which indicates that t2 = {(x2 /1)/(w/n)}, where w
2 (n) and x2 2 (1), since x N (0, 1).
11
D.S.G. POLLOCK: ECONOMETRICS
(46) Let N (0, Q), where null(Q) = 0. Then there exists a nonsin-
gular matrix T such that T 0 T = Q1 , T QT 0 = I, and it follows
that T N (0, I).
This result can be used immediately to prove the first result concerning
quadratic forms:
12