ADVANCED DISCRETE DISTRIBUTIONS
7.1 Compound Frequency Distributions
7
π=π1+π2+β―+ππ(whereπ = 0implies thatπ= 0) isππ(π§) =ππ[ππ(π§)]. This is shown as follows:
ππ(π§) =
ββ π=0
Pr(π=π)π§π=
ββ π=0
ββ π=0
Pr(π=π|π=π) Pr(π=π)π§π
=
ββ π=0
Pr(π=π)
ββ π=0
Pr(π1+β―+ππ=π|π=π)π§π
=
ββ π=0
Pr(π=π)[ππ(π§)]π
=ππ[ππ(π§)].
In insurance contexts, this distribution can arise naturally. Ifπrepresents the number of accidents arising in a portfolio of risks and{ππβΆ π= 1,2,β¦, π}represents the number of claims (injuries, number of cars, etc.) from the accidents, thenπ represents the total number of claims from the portfolio. This kind of interpretation is not necessary to justify the use of a compound distribution. If a compound distribution fits data well, that may be enough justification itself. Also, there are other motivations for these distributions, as presented in Section 7.5.
EXAMPLE 7.1
Demonstrate that any zero-modified distribution is a compound distribution.
Consider a primary Bernoulli distribution. It has pgfππ(π§) = 1 βπ+ππ§. Then consider an arbitrary secondary distribution with pgfππ(π§). Then, from (7.1), we obtain
ππ(π§) =ππ[ππ(π§)] = 1 βπ+πππ(π§). From (6.4) this is the pgf of a ZM distribution with
π= 1 βππ0 1 βπ0
.
That is, the ZM distribution has assigned arbitrary probabilityππ0 at zero, whileπ0is the probability assigned at zero by the secondary distribution. β‘ EXAMPLE 7.2
Consider the case where bothπ andπhave a Poisson distribution. Determine the pgf of this distribution.
This distribution is called the PoissonβPoisson or Neyman Type A distribution.
Letππ(π§) =ππ1(π§β1)andππ(π§) =ππ2(π§β1). Then, ππ(π§) =ππ1[ππ2(π§β1)β1].
Whenπ2is a lot larger thanπ1, for example, π1 = 0.1 andπ2 = 10, the resulting
distribution will have two local modes. β‘
The probability of exactlyπclaims can be written as Pr(π=π) =
ββ π=0
Pr(π=π|π=π) Pr(π=π)
=
ββ π=0
Pr(π1+β―+ππ =π|π=π) Pr(π=π)
=
ββ π=0
Pr(π1+β―+ππ=π) Pr(π=π). (7.2) Lettingππ= Pr(π=π),ππ= Pr(π=π), andππ= Pr(π =π), this is rewritten as
ππ=
ββ π=0
ππππβπ, (7.3)
whereππβπ, π= 0,1,β¦, is the βπ-fold convolutionβ of the functionππ, π= 0,1,β¦, that is, the probability that the sum ofπrandom variables which are each i.i.d. with probability functionππwill take on valueπ.
Whenππ(π§)is chosen to be a member of the(π, π,0)class, ππ=
(π+ π π
)ππβ1, π= 1,2,β¦, (7.4)
then a simple recursive formula can be used. This formula avoids the use of convolutions and thus reduces the computations considerably.
Theorem 7.1 If the primary distribution is a member of the(π, π,0)class, the recursive formula is
ππ= 1 1 βππ0
βπ π=1
( π+ππ
π )
ππππβπ, π= 1,2,3,β¦. (7.5)
Proof:From (7.4),
πππ=π(πβ 1)ππβ1+ (π+π)ππβ1.
Multiplying each side by[ππ(π§)]πβ1ππβ² (π§)and summing overπyields
ββ π=1
πππ[ππ(π§)]πβ1ππβ² (π§) =π
ββ π=1
(πβ 1)ππβ1[ππ(π§)]πβ1ππβ² (π§)
+ (π+π)
ββ π=1
ππβ1[ππ(π§)]πβ1ππβ² (π§).
Becauseππ(π§) =ββ
π=0ππ[ππ(π§)]π, the previous equation is ππβ²(π§) =π
ββ π=0
πππ[ππ(π§)]πππβ² (π§) + (π+π)
ββ π=0
ππ[ππ(π§)]πππβ² (π§).
Therefore,
ππβ²(π§) =πππβ²(π§)ππ(π§) + (π+π)ππ(π§)ππβ² (π§).
Each side can be expanded in powers ofπ§. The coefficients ofπ§πβ1in such an expansion must be the same on both sides of the equation. Hence, forπ= 1,2,β¦, we have
πππ=π
βπ π=0
(πβπ)ππππβπ + (π+π)
βπ π=0
πππππβπ
=πππ0ππ+πβπ
π=1
(πβπ)ππππβπ+ (π+π)
βπ π=1
πππππβπ
=πππ0ππ+ππβπ
π=1
ππππβπ+πβπ
π=1
πππππβπ. Therefore,
ππ=ππ0ππ+
βπ π=1
( π+ππ
π )
ππππβπ.
Rearrangement yields (7.5). β‘
In order to use (7.5), the starting valueπ0is required and is given in Theorem 7.3. If the primary distribution is a member of the(π, π,1) class, the proof must be modified to reflect the fact that the recursion for the primary distribution begins atπ= 2. The result is the following.
Theorem 7.2 If the primary distribution is a member of the(π, π,1)class, the recursive formula is
ππ=
[π1β (π+π)π0]ππ+βπ
π=1(π+ππβπ)ππππβπ
1 βππ0
, π= 1,2,3,β¦. (7.6)
Proof:It is similar to the proof of Theorem 7.1 and is omitted. β‘ EXAMPLE 7.3
Develop the recursive formula for the case where the primary distribution is Poisson.
In this case,π= 0andπ=π, yielding the recursive form ππ= π
π
βπ π=1
πππππβπ. The starting value is, from (7.1),
π0= Pr(π= 0) =π(0)
=ππ[ππ(0)] =ππ(π0)
=πβπ(1βπ0). (7.7)
Distributions of this type are calledcompound Poisson. When the secondary distribu- tion is specified, the compound distribution is called PoissonβX, where X is the name
of the secondary distribution. β‘
The method used to obtainπ0applies to any compound distribution.
Theorem 7.3 For any compound distribution,π0=ππ(π0), whereππ(π§)is the pgf of the primary distribution andπ0is the probability that the secondary distribution takes on the value zero.
Proof:See the second line of (7.7) β‘
Note that the secondary distribution is not required to be in any special form. However, to keep the number of distributions manageable, secondary distributions are selected from the(π, π,0)or the(π, π,1)class.
EXAMPLE 7.4
Calculate the probabilities for the PoissonβETNB distribution, whereπ = 3for the Poisson distribution and the ETNB distribution hasπ= β0.5andπ½ = 1.
From Example 6.5, the secondary probabilities are π0 = 0, π1 = 0.853553, π2 = 0.106694, andπ3 = 0.026674. From (7.7),π0= exp[β3(1 β 0)] = 0.049787.
For the Poisson primary distribution,π= 0andπ= 3. The recursive formula in (7.5) becomes
ππ=
βπ
π=1(3πβπ)ππππβπ
1 β 0(0) =
βπ π=1
3π πππππβπ. Then,
π1= 3(1)
1 0.853553(0.049787) = 0.127488, π2= 3(1)
2 0.853553(0.127488) +3(2)
2 0.106694(0.049787) = 0.179163, π3= 3(1)
3 0.853553(0.179163) +3(2)
3 0.106694(0.127488) +3(3)
3 0.026674(0.049787) = 0.184114. β‘
EXAMPLE 7.5
Demonstrate that the Poissonβlogarithmic distribution is a negative binomial distribution.
The negative binomial distribution has pgf π(π§) = [1 βπ½(π§β 1)]βπ.
Suppose thatππ(π§)is Poisson(π)andππ(π§)is logarithmic(π½). Then, ππ[ππ(π§)] = exp{π[ππ(π§) β 1]}
= exp {
π [
1 βln[1 βπ½(π§β 1)]
ln(1 +π½) β 1 ]}
= exp
{ βπ
ln(1 +π½)ln[1 βπ½(π§β 1)]
}
= [1 βπ½(π§β 1)]βπβ ln(1+π½)
= [1 βπ½(π§β 1)]βπ,
whereπ = πβ ln(1 +π½). This shows that the negative binomial distribution can be written as a compound Poisson distribution with a logarithmic secondary distribution.
β‘ Example 7.5 shows that the Poissonβlogarithmic distribution does not create a new distri- bution beyond the(π, π,0)and(π, π,1)classes. As a result, this combination of distributions is not useful to us. Another combination that does not create a new distribution beyond the (π, π,1) class is the compound geometric distribution, where both the primary and secondary distributions are geometric. The resulting distribution is a zero-modified geometric distribution, as shown in Exercise 7.2. The following theorem shows that certain other combinations are also of no use in expanding the class of distributions through compounding. Suppose thatππ(π§) = ππ[ππ(π§)]as before. Now,ππ(π§)can always be written as
ππ(π§) =π0+ (1 βπ0)ππβ(π§), (7.8) whereππβ(π§)is the pgf of the conditional distribution over the positive range (in other words, the zero-truncated version).
Theorem 7.4 Suppose that the pgfππ(π§;π)satisfies ππ(π§;π) =π΅[π(π§β 1)]
for some parameter π and some function π΅(π§) that is independent of π. That is, the parameterπand the argumentπ§only appear in the pgf asπ(π§β 1). There may be other parameters as well, and they may appear anywhere in the pgf. Then,ππ(π§) =ππ[ππ(π§);π] can be rewritten as
ππ(π§) =ππ[πππ(π§);π(1 βπ0)].
Proof:
ππ(π§) =ππ[ππ(π§);π]
=ππ[π0+ (1 βπ0)πππ(π§);π]
=π΅{π[π0+ (1 βπ0)πππ(π§) β 1]}
=π΅{π(1 βπ0)[πππ(π§) β 1]}
=ππ[πππ(π§);π(1 βπ0)]. β‘
This shows that adding, deleting, or modifying the probability at zero in the secondary distribution does not add a new distribution because it is equivalent to modifying the parameterπof the primary distribution. Thus, for example, a Poisson primary distribution with a Poisson, zero-truncated Poisson, or zero-modified Poisson secondary distribution will still lead to a Neyman Type A (PoissonβPoisson) distribution.
EXAMPLE 7.6
Determine the probabilities for a Poissonβzero-modified ETNB distribution where the parameters areπ= 7.5,ππ0 = 0.6,π= β0.5, andπ½= 1.
From Example 6.5, the secondary probabilities are π0 = 0.6, π1 = 0.341421, π2= 0.042678, andπ3= 0.010670. From (7.7),π0= exp[β7.5(1 β 0.6)] = 0.049787.
For the Poisson primary distribution, π= 0andπ = 7.5. The recursive formula in (7.5) becomes
ππ=
βπ
π=1(7.5πβπ)ππππβπ
1 β 0(0.6) =
βπ π=1
7.5π π ππππβπ. Then,
π1= 7.5(1)
1 0.341421(0.049787) = 0.127487, π2= 7.5(1)
2 0.341421(0.127487) +7.5(2)
2 0.042678(0.049787) = 0.179161, π3= 7.5(1)
3 0.341421(0.179161) +7.5(2)
3 0.042678(0.127487) + 7.5(3)
3 0.010670(0.049787) = 0.184112.
Except for slight rounding differences, these probabilities are the same as those
obtained in Example 7.4. β‘
7.1.1 Exercises
7.1 Do all the members of the(π, π,0)class satisfy the condition of Theorem 7.4? For those that do, identify the parameter (or function of its parameters) that plays the role ofπ in the theorem.
7.2 Show that the following three distributions are identical: (1) geometricβgeometric, (2) Bernoulliβgeometric, and (3) zero-modified geometric. That is, for any one of the distributions with arbitrary parameters, show that there is a member of the other two distribution types that has the same pf or pgf.
7.3 Show that the binomialβgeometric and negative binomialβgeometric (with negative binomial parameterπa positive integer) distributions are identical.