30.7 GENERATING FUNCTIONS
Comparing this expression with (30.92), we find thatκ 1 =μ,κ 2 =σ^2 and all other
cumulants are equal to zero.
We may obtain expressions for the cumulants of a distribution in terms of its
moments by differentiating (30.92) with respect totto give
dKX
dt
=
1
MX
dMX
dt
.
Expanding each term as power series intand cross-multiplying, we obtain
(
κ 1 +κ 2 t+κ 3
t^2
2!
+···
)(
1+μ 1 t+μ 2
t^2
2!
+···
)
=
(
μ 1 +μ 2 t+μ 3
t^2
2!
+···
)
,
and, on equating coefficients of like powers ofton each side, we find
μ 1 =κ 1 ,
μ 2 =κ 2 +κ 1 μ 1 ,
μ 3 =κ 3 +2κ 2 μ 1 +κ 1 μ 2 ,
μ 4 =κ 4 +3κ 3 μ 1 +3κ 2 μ 2 +κ 1 μ 3 ,
..
.
μk=κk+k−^1 C 1 κk− 1 μ 1 +···+k−^1 Crκk−rμr+···+κ 1 μk− 1.
Solving these equations for theκk, we obtain (for the first four cumulants)
κ 1 =μ 1 ,
κ 2 =μ 2 −μ^21 =ν 2 ,
κ 3 =μ 3 − 3 μ 2 μ 1 +2μ^31 =ν 3 ,
κ 4 =μ 4 − 4 μ 3 μ 1 +12μ 2 μ^21 − 3 μ^22 − 6 μ^41 =ν 4 − 3 ν^22. (30.93)
Higher-order cumulants may be calculated in the same way but become increas-
ingly lengthy to write out in full.
The principal property of cumulants is their additivity, which may be proved
by combining (30.92) with (30.90). IfX 1 ,X 2 , ...,XNare independent random
variables andKXi(t)fori=1, 2 ,...,Nis the CGF forXithen the CGF of
SN=c 1 X 1 +c 2 X 2 +···+cNXN(where theciare constants) is given by
KSN(t)=
∑N
i=1
KXi(cit).
Cumulants also have the useful property that, under a change of originX→
X+athe first cumulant undergoes the changeκ 1 →κ 1 +abut all higher-order
cumulants remain unchanged. Under a change of scaleX→bX, cumulantκr
undergoes the changeκr→brκr.