Tables for

Exam STAM
 The reading material for Exam STAM includes a variety of textbooks.
Each text has a set of probability distributions that are used in its readings.
For those distributions used in more than one text, the choices of
parameterization may not be the same in all of the books. This may be of
educational value while you study, but could add a layer of uncertainty in
the examination. For this latter reason, we have adopted one set of
parameterizations to be used in examinations. This set will be based
on Appendices A & B of Loss Models: From Data to Decisions by
Klugman, Panjer and Willmot. A slightly revised version of these
appendices is included in this note. A copy of this note will also be distributed
to each candidate at the examination.

Each text also has its own system of dedicated notation and
terminology. Sometimes these may conflict. If alternative meanings could
apply in an examination question, the symbols will be defined.

For Exam STAM, in addition to the abridged table from Loss Models,
sets of values from the standard normal and chi-square distributions
will be available for use in examinations. These are also included in this note.

When using the normal distribution, choose the nearest z-value to find the
probability, or if the probability is given, choose the nearest z-value. No
interpolation should be used.

Example: If the given z-value is 0.759, and you need to find Pr(Z < 0.759) from
the normal distribution table, then choose the probability for z-value = 0.76: Pr(Z
< 0.76) = 0.7764.

When using the normal approximation to a discrete distribution, use the

continuity correction.

1 − 12 x2
The density function for the standard normal distribution is φ ( x) = e .
2π
Excerpts from the Appendices to Loss Models: From Data to
Decisions, 3rd edition

August 27, 2009

Appendix A

An Inventory of Continuous
Distributions

A.1 Introduction
The incomplete gamma function is given by
Z x
1
Γ(α; x) = tα−1 e−t dt, α > 0, x > 0
Γ(α) 0
Z ∞
with Γ(α) = tα−1 e−t dt, α > 0.
0
Also, define Z ∞
G(α; x) = tα−1 e−t dt, x > 0.
x
At times we will need this integral for nonpositive values of α. Integration by parts produces the relationship

xα e−x 1
G(α; x) = − + G(α + 1; x)
α α
This can be repeated until the first argument of G is α + k, a positive number. Then it can be evaluated
from
G(α + k; x) = Γ(α + k)[1 − Γ(α + k; x)].
The incomplete beta function is given by
Z
Γ(a + b) x a−1
β(a, b; x) = t (1 − t)b−1 dt, a > 0, b > 0, 0 < x < 1.
Γ(a)Γ(b) 0

1
APPENDIX A. AN INVENTORY OF CONTINUOUS DISTRIBUTIONS 2

A.2 Transformed beta family

A.2.2 Three-parameter distributions
A.2.2.1 Generalized Pareto (beta of the second kind)–α, θ, τ

Γ(α + τ ) θα xτ −1 x
f (x) = F (x) = β(τ , α; u), u=
Γ(α)Γ(τ ) (x + θ)α+τ x+θ
θk Γ(τ + k)Γ(α − k)
E[X k ] = , −τ < k < α
Γ(α)Γ(τ )
θk τ (τ + 1) · · · (τ + k − 1)
E[X k ] = , if k is an integer
(α − 1) · · · (α − k)
θk Γ(τ + k)Γ(α − k)
E[(X ∧ x)k ] = β(τ + k, α − k; u) + xk [1 − F (x)], k > −τ
Γ(α)Γ(τ )
τ −1
mode = θ , τ > 1, else 0
α+1

A.2.2.2 Burr (Burr Type XII, Singh-Maddala)–α, θ, γ

αγ(x/θ)γ 1
f (x) = F (x) = 1 − uα , u=
x[1 + (x/θ)γ ]α+1 1 + (x/θ)γ
θk Γ(1 + k/γ)Γ(α − k/γ)
E[X k ] = , −γ < k < αγ
Γ(α)
VaRp (X) = θ[(1 − p)−1/α − 1]1/γ
θk Γ(1 + k/γ)Γ(α − k/γ)
E[(X ∧ x)k ] = β(1 + k/γ, α − k/γ; 1 − u) + xk uα , k > −γ
Γ(α)
µ ¶1/γ
γ−1
mode = θ , γ > 1, else 0
αγ + 1

A.2.2.3 Inverse Burr (Dagum)–τ , θ , γ

τ γ(x/θ)γτ (x/θ)γ
f (x) = F (x) = uτ , u=
x[1 + (x/θ)γ ]τ +1 1 + (x/θ)γ
θk Γ(τ + k/γ)Γ(1 − k/γ)
E[X k ] = , −τ γ < k < γ
Γ(τ )
VaRp (X) = θ(p−1/τ − 1)−1/γ
θk Γ(τ + k/γ)Γ(1 − k/γ)
E[(X ∧ x)k ] = β(τ + k/γ, 1 − k/γ; u) + xk [1 − uτ ], k > −τ γ
Γ(τ )
µ ¶1/γ
τγ − 1
mode = θ , τ γ > 1, else 0
γ+1
APPENDIX A. AN INVENTORY OF CONTINUOUS DISTRIBUTIONS 3

A.2.3 Two-parameter distributions

A.2.3.1 Pareto (Pareto Type II, Lomax)–α, θ
µ ¶α
αθα θ
f (x) = F (x) = 1 −
(x + θ)α+1 x+θ
θk Γ(k + 1)Γ(α − k)
E[X k ] = , −1 < k < α
Γ(α)
θk k!
E[X k ] = , if k is an integer
(α − 1) · · · (α − k)
VaRp (X) = θ[(1 − p)−1/α − 1]
θ(1 − p)−1/α
TVaRp (X) = VaRp (X) + , α>1
α−1
" µ ¶α−1 #
θ θ
E[X ∧ x] = 1− , α=
6 1
α−1 x+θ
µ ¶
θ
E[X ∧ x] = −θ ln , α=1
x+θ
µ ¶α
θk Γ(k + 1)Γ(α − k) θ
E[(X ∧ x)k ] = β[k + 1, α − k; x/(x + θ)] + xk , all k
Γ(α) x+θ
mode = 0

A.2.3.2 Inverse Pareto–τ , θ

µ ¶τ
τ θxτ −1 x
f (x) = F (x) =
(x + θ)τ +1 x+θ
θk Γ(τ + k)Γ(1 − k)
E[X k ] = , −τ < k < 1
Γ(τ )
θk (−k)!
E[X k ] = , if k is a negative integer
(τ − 1) · · · (τ + k)
VaRp (X) = θ[p−1/τ − 1]−1
Z x/(x+θ) ∙ µ ¶τ ¸
x
E[(X ∧ x)k ] = θk τ y τ +k−1 (1 − y)−k dy + xk 1 − , k > −τ
0 x+θ
τ −1
mode = θ , τ > 1, else 0
2

A.2.3.3 Loglogistic (Fisk)–γ, θ

γ(x/θ)γ (x/θ)γ
f (x) = F (x) = u, u=
x[1 + (x/θ)γ ]2 1 + (x/θ)γ
E[X k ] = θk Γ(1 + k/γ)Γ(1 − k/γ), −γ < k < γ
VaRp (X) = θ(p−1 − 1)−1/γ
E[(X ∧ x)k ] = θk Γ(1 + k/γ)Γ(1 − k/γ)β(1 + k/γ, 1 − k/γ; u) + xk (1 − u), k > −γ
µ ¶1/γ
γ−1
mode = θ , γ > 1, else 0
γ+1
APPENDIX A. AN INVENTORY OF CONTINUOUS DISTRIBUTIONS 4

A.2.3.4 Paralogistic–α, θ
This is a Burr distribution with γ = α.

α2 (x/θ)α 1
f (x) = F (x) = 1 − uα , u=
x[1 + (x/θ)α ]α+1 1 + (x/θ)α
θk Γ(1 + k/α)Γ(α − k/α)
E[X k ] = , −α < k < α2
Γ(α)
VaRp (X) = θ[(1 − p)−1/α − 1]1/α
θk Γ(1 + k/α)Γ(α − k/α)
E[(X ∧ x)k ] = β(1 + k/α, α − k/α; 1 − u) + xk uα , k > −α
Γ(α)
µ ¶1/α
α−1
mode = θ , α > 1, else 0
α2 + 1

A.2.3.5 Inverse paralogistic–τ , θ

This is an inverse Burr distribution with γ = τ .

2
τ 2 (x/θ)τ (x/θ)τ
f (x) = F (x) = uτ , u=
x[1 + (x/θ)τ ]τ +1 1 + (x/θ)τ
θk Γ(τ + k/τ )Γ(1 − k/τ )
E[X k ] = , −τ 2 < k < τ
Γ(τ )
VaRp (X) = θ(p−1/τ − 1)−1/τ
θk Γ(τ + k/τ )Γ(1 − k/τ )
E[(X ∧ x)k ] = β(τ + k/τ , 1 − k/τ ; u) + xk [1 − uτ ], k > −τ 2
Γ(τ )
mode = θ (τ − 1)1/τ , τ > 1, else 0

A.3 Transformed gamma family

A.3.2 Two-parameter distributions
A.3.2.1 Gamma–α, θ

(x/θ)α e−x/θ
f (x) = F (x) = Γ(α; x/θ)
xΓ(α)
θk Γ(α + k)
M (t) = (1 − θt)−α , t < 1/θ E[X k ] = , k > −α
Γ(α)
E[X k ] = θk (α + k − 1) · · · α, if k is an integer

θk Γ(α + k)
E[(X ∧ x)k ] = Γ(α + k; x/θ) + xk [1 − Γ(α; x/θ)], k > −α
Γ(α)
= α(α + 1) · · · (α + k − 1)θk Γ(α + k; x/θ) + xk [1 − Γ(α; x/θ)], k an integer
mode = θ(α − 1), α > 1, else 0
APPENDIX A. AN INVENTORY OF CONTINUOUS DISTRIBUTIONS 5

A.3.2.2 Inverse gamma (Vinci)–α, θ

(θ/x)α e−θ/x
f (x) = F (x) = 1 − Γ(α; θ/x)
xΓ(α)
θk Γ(α − k) θk
E[X k ] = , k<α E[X k ] = , if k is an integer
Γ(α) (α − 1) · · · (α − k)
θk Γ(α − k)
E[(X ∧ x)k ] = [1 − Γ(α − k; θ/x)] + xk Γ(α; θ/x)
Γ(α)
θk Γ(α − k)
= G(α − k; θ/x) + xk Γ(α; θ/x), all k
Γ(α)
mode = θ/(α + 1)

A.3.2.3 Weibull–θ, τ

τ
τ (x/θ)τ e−(x/θ) τ
f (x) = F (x) = 1 − e−(x/θ)
x
E[X k ] = θk Γ(1 + k/τ ), k > −τ
VaRp (X) = θ[− ln(1 − p)]1/τ
τ
E[(X ∧ x)k ] = θk Γ(1 + k/τ )Γ[1 + k/τ ; (x/θ)τ ] + xk e−(x/θ) , k > −τ
µ ¶1/τ
τ −1
mode = θ , τ > 1, else 0
τ

A.3.2.4 Inverse Weibull (log Gompertz)–θ, τ

τ
τ (θ/x)τ e−(θ/x) τ
f (x) = F (x) = e−(θ/x)
x
E[X k ] = θk Γ(1 − k/τ ), k < τ
VaRp (X) = θ(− ln p)−1/τ
h τ
i
E[(X ∧ x)k ] = θk Γ(1 − k/τ ){1 − Γ[1 − k/τ ; (θ/x)τ ]} + xk 1 − e−(θ/x) , all k
h τ
i
= θk Γ(1 − k/τ )G[1 − k/τ ; (θ/x)τ ] + xk 1 − e−(θ/x)
µ ¶1/τ
τ
mode = θ
τ +1
APPENDIX A. AN INVENTORY OF CONTINUOUS DISTRIBUTIONS 6

A.3.3 One-parameter distributions

A.3.3.1 Exponential–θ

e−x/θ
f (x) = F (x) = 1 − e−x/θ
θ
M (t) = (1 − θt)−1 E[X k ] = θk Γ(k + 1), k > −1
k
E[X k ] = θ k!, if k is an integer
VaRp (X) = −θ ln(1 − p)
TVaRp (X) = −θ ln(1 − p) + θ
E[X ∧ x] = θ(1 − e−x/θ )
E[(X ∧ x)k ] = θk Γ(k + 1)Γ(k + 1; x/θ) + xk e−x/θ , k > −1
= θk k!Γ(k + 1; x/θ) + xk e−x/θ , k an integer
mode = 0

A.3.3.2 Inverse exponential–θ

θe−θ/x
f (x) = F (x) = e−θ/x
x2
k k
E[X ] = θ Γ(1 − k), k<1
VaRp (X) = θ(− ln p)−1
E[(X ∧ x)k ] = θk G(1 − k; θ/x) + xk (1 − e−θ/x ), all k
mode = θ/2

A.5 Other distributions

A.5.1.1 Lognormal–μ,σ (μ can be negative)

1 ln x − μ
f (x) = √ exp(−z 2 /2) = φ(z)/(σx), z = F (x) = Φ(z)
xσ 2π σ
k 2 2
E[X ] = exp(kμ + k σ /2)
µ ¶
k 2 2 ln x − μ − kσ 2
E[(X ∧ x) ] = exp(kμ + k σ /2)Φ + xk [1 − F (x)]
σ
mode = exp(μ − σ2 )
APPENDIX A. AN INVENTORY OF CONTINUOUS DISTRIBUTIONS 7

A.5.1.2 Inverse Gaussian–μ, θ

µ ¶1/2 µ ¶
θ θz 2 x−μ
f (x) = exp − , z=
2πx3 2x μ
" µ ¶ # µ ¶ " µ ¶1/2 #
1/2
θ 2θ θ x+μ
F (x) = Φ z + exp Φ −y , y=
x μ x μ
" Ã r !#
θ 2tμ2 θ
M (t) = exp 1− 1− , t < 2, E[X] = μ, Var[X] = μ3 /θ
μ θ 2μ
" µ ¶ # µ ¶ " µ ¶1/2 #
1/2
θ 2θ θ
E[X ∧ x] = x − μzΦ z − μy exp Φ −y
x μ x

A.5.1.3 log-t–r, μ, σ (μ can be negative)

Let Y have a t distribution with r degrees of freedom. Then X = exp(σY + μ) has the log-t distribution.
Positive moments do not exist for this distribution. Just as the t distribution has a heavier tail than the
normal distribution, this distribution has a heavier tail than the lognormal distribution.
µ ¶
r+1
Γ
2
f (x) = " ,
³r´ µ ¶2 #(r+1)/2
√ 1 ln x − μ
xσ πrΓ 1+
2 r σ
µ ¶
ln x − μ
F (x) = Fr with Fr (t) the cdf of a t distribution with r d.f.,
σ
⎧ ⎡ ⎤
⎪
⎪
⎪
⎪
⎪
⎪
⎪ 1 ⎢ ⎢r 1 r ⎥
⎥
⎪ β ⎢
⎪ 2 ⎣2 2 , ; µ ¶2⎥ , 0 < x ≤ eμ ,
⎪
⎪ ln x − μ ⎦
⎪
⎨ r+
⎡ σ ⎤
F (x) =
⎪
⎪
⎪
⎪
⎪
⎪
⎪ 1 ⎢ ⎢r 1 r ⎥
⎥
⎪
⎪ 1 − β ⎢ , ; µ ¶2 ⎥ , x ≥ eμ .
⎪
⎪ 2 ⎣ 2 2 ln x − μ ⎦
⎪
⎩ r+
σ

A.5.1.4 Single-parameter Pareto–α, θ

αθα
f (x) = , x>θ F (x) = 1 − (θ/x)α , x>θ
xα+1
αθ(1 − p)−1/α
VaRp (X) = θ(1 − p)−1/α TVaRp (X) = , α>1
α−1
αθk αθk kθα
E[X k ] = , k<α E[(X ∧ x)k ] = − , x≥θ
α−k α − k (α − k)xα−k
mode = θ

Note: Although there appears to be two parameters, only α is a true parameter. The value of θ must be
set in advance.
APPENDIX A. AN INVENTORY OF CONTINUOUS DISTRIBUTIONS 8

A.6 Distributions with finite support

For these two distributions, the scale parameter θ is assumed known.

A.6.1.1 Generalized beta–a, b, θ, τ

Γ(a + b) a τ
f (x) = u (1 − u)b−1 , 0 < x < θ, u = (x/θ)τ
Γ(a)Γ(b) x
F (x) = β(a, b; u)
θk Γ(a + b)Γ(a + k/τ )
E[X k ] = , k > −aτ
Γ(a)Γ(a + b + k/τ )
θk Γ(a + b)Γ(a + k/τ )
E[(X ∧ x)k ] = β(a + k/τ , b; u) + xk [1 − β(a, b; u)]
Γ(a)Γ(a + b + k/τ )

A.6.1.2 beta–a, b, θ

Γ(a + b) a 1
f (x) = u (1 − u)b−1 , 0 < x < θ, u = x/θ
Γ(a)Γ(b) x
F (x) = β(a, b; u)
θk Γ(a + b)Γ(a + k)
E[X k ] = , k > −a
Γ(a)Γ(a + b + k)
θk a(a + 1) · · · (a + k − 1)
E[X k ] = , if k is an integer
(a + b)(a + b + 1) · · · (a + b + k − 1)
θk a(a + 1) · · · (a + k − 1)
E[(X ∧ x)k ] = β(a + k, b; u)
(a + b)(a + b + 1) · · · (a + b + k − 1)
+xk [1 − β(a, b; u)]
Appendix B

An Inventory of Discrete
Distributions

B.1 Introduction
The 16 models fall into three classes. The divisions are based on the algorithm by which the probabilities are
computed. For some of the more familiar distributions these formulas will look different from the ones you
may have learned, but they produce the same probabilities. After each name, the parameters are given. All
parameters are positive unless otherwise indicated. In all cases, pk is the probability of observing k losses.
For finding moments, the most convenient form is to give the factorial moments. The jth factorial
moment is μ(j) = E[N (N − 1) · · · (N − j + 1)]. We have E[N ] = μ(1) and Var(N ) = μ(2) + μ(1) − μ2(1) .
The estimators which are presented are not intended to be useful estimators but rather for providing
starting values for maximizing the likelihood (or other) function. For determining starting values, the
following quantities are used [where nk is the observed frequency at k (if, for the last entry, nk represents
the number of observations at k or more, assume it was at exactly k) and n is the sample size]:
∞ ∞
1X 1X 2
μ̂ = knk , σ̂ 2 = k nk − μ̂2 .
n n
k=1 k=1

When the method of moments is used to determine the starting value, a circumflex (e.g., λ̂) is used. For
any other method, a tilde (e.g., λ̃) is used. When the starting value formulas do not provide admissible
parameter values, a truly crude guess is to set the product of all λ and β parameters equal to the sample
mean and set all other parameters equal to 1. If there are two λ and/or β parameters, an easy choice is to
set each to the square root of the sample mean.
The last item presented is the probability generating function,

P (z) = E[z N ].

B.2 The (a, b, 0) class

B.2.1.1 Poisson–λ

e−λ λk
p0 = e−λ , a = 0, b=λ pk =
k!
E[N ] = λ, Var[N ] = λ P (z) = eλ(z−1)

9
APPENDIX B. AN INVENTORY OF DISCRETE DISTRIBUTIONS 10

B.2.1.2 Geometric–β

1 β βk
p0 = , a= , b=0 pk =
1+β 1+β (1 + β)k+1
E[N ] = β, Var[N ] = β(1 + β) P (z) = [1 − β(z − 1)]−1 .

This is a special case of the negative binomial with r = 1.

B.2.1.3 Binomial–q, m, (0 < q < 1, m an integer)

q (m + 1)q
p0= (1 − q)m , a = − , b=
1−q 1−q
µ ¶
m k
pk = q (1 − q)m−k , k = 0, 1, . . . , m
k
E[N ] = mq, Var[N ] = mq(1 − q) P (z) = [1 + q(z − 1)]m .

B.2.1.4 Negative binomial–β, r

β (r − 1)β
p0 = (1 + β)−r , a= , b=
1+β 1+β
r(r + 1) · · · (r + k − 1)β k
pk =
k!(1 + β)r+k
E[N ] = rβ, Var[N ] = rβ(1 + β) P (z) = [1 − β(z − 1)]−r .

B.3 The (a, b, 1) class

To distinguish this class from the (a, b, 0) class, the probabilities are denoted Pr(N = k) = pM k or Pr(N =
k) = pTk depending on which subclass is being represented. For this class, pM 0 is arbitrary (that is, it is a
parameter) and then pM T
1 or p1 is a specified function of the parameters a and b. Subsequent probabilities are
obtained recursively as in the (a, b, 0) class: pM M
k = (a + b/k)pk−1 , k = 2, 3, . . ., with the same recursion for pk
T

There are two sub-classes of this class. When discussing their members, we often refer to the “corresponding”
member of the (a, b, 0) class. This refers to the member of that class with the same values for a and b. The
notation pk will continue to be used for probabilities for the corresponding (a, b, 0) distribution.

B.3.1 The zero-truncated subclass

The members of this class have pT0 = 0 and therefore it need not be estimated. These distributions should
only be used when a value of zero is impossible. The first factorial moment is μ(1) = (a + b)/[(1 − a)(1 − p0 )],
where p0 is the value for the corresponding member of the (a, b, 0) class. For the logarithmic distribution
(which has no corresponding member), μ(1) = β/ ln(1+β). Higher factorial moments are obtained recursively
with the same formula as with the (a, b, 0) class. The variance is (a+b)[1−(a+b+1)p0 ]/[(1−a)(1−p0 )]2 .For
those members of the subclass which have corresponding (a, b, 0) distributions, pTk = pk /(1 − p0 ).
APPENDIX B. AN INVENTORY OF DISCRETE DISTRIBUTIONS 11

B.3.1.1 Zero-truncated Poisson–λ

λ
pT1 = , a = 0, b = λ,
eλ − 1
λk
pTk = ,
k!(eλ − 1)
E[N ] = λ/(1 − e−λ ), Var[N ] = λ[1 − (λ + 1)e−λ ]/(1 − e−λ )2 ,
λ̃ = ln(nμ̂/n1 ),
eλz − 1
P (z) = .
eλ − 1

B.3.1.2 Zero-truncated geometric–β

1 β
pT1 = , a= , b = 0,
1+β 1+β
β k−1
pTk = ,
(1 + β)k
E[N ] = 1 + β, Var[N ] = β(1 + β),
β̂ = μ̂ − 1,
[1 − β(z − 1)]−1 − (1 + β)−1
P (z) = .
1 − (1 + β)−1

This is a special case of the zero-truncated negative binomial with r = 1.

B.3.1.3 Logarithmic–β

β β β
pT1 = , a= , b=− ,
(1 + β) ln(1 + β) 1+β 1+β
βk
pTk = ,
k(1 + β)k ln(1 + β)
β[1 + β − β/ ln(1 + β)]
E[N ] = β/ ln(1 + β), Var[N ] = ,
ln(1 + β)
nμ̂ 2(μ̂ − 1)
β̃ =− 1 or ,
n1 μ̂
ln[1 − β(z − 1)]
P (z) = 1 − .
ln(1 + β)

This is a limiting case of the zero-truncated negative binomial as r → 0.

APPENDIX B. AN INVENTORY OF DISCRETE DISTRIBUTIONS 12

B.3.1.4 Zero-truncated binomial–q, m, (0 < q < 1, m an integer)

m(1 − q)m−1 q q (m + 1)q

pT1 = m
, a=− , b= ,
1 − (1 − q) 1−q 1−q
¡m¢ k m−k
k q (1 − q)
pTk = , k = 1, 2, . . . , m,
1 − (1 − q)m
mq
E[N ] = ,
1 − (1 − q)m
mq[(1 − q) − (1 − q + mq)(1 − q)m ]
Var[N ] = ,
[1 − (1 − q)m ]2
μ̂
q̃ = ,
m
[1 + q(z − 1)]m − (1 − q)m
P (z) = .
1 − (1 − q)m

B.3.1.5 Zero-truncated negative binomial–β, r, (r > −1, r 6= 0)

rβ β (r − 1)β
pT1 = , a= , b= ,
(1 +β)r+1 − (1 + β) 1+β 1+β
µ ¶k
r(r + 1) · · · (r + k − 1) β
pTk = ,
k![(1 + β)r − 1] 1+β
rβ
E[N ] = ,
1 − (1 + β)−r
rβ[(1 + β) − (1 + β + rβ)(1 + β)−r ]
V ar[N ] = ,
[1 − (1 + β)−r ]2
σ̂2 μ̂2
β̃ = − 1, r̃ = 2 ,
μ̂ σ̂ − μ̂
[1 − β(z − 1)]−r − (1 + β)−r
P (z) = .
1 − (1 + β)−r

This distribution is sometimes called the extended truncated negative binomial distribution because the
parameter r can extend below 0.

B.3.2 The zero-modified subclass

A zero-modified distribution is created by starting with a truncated distribution and then placing an arbitrary
amount of probability at zero. This probability, pM 0 , is a parameter. The remaining probabilities are
adjusted accordingly. Values of pM k can be determined from the corresponding zero-truncated distribution
as pM
k = (1 − pM T
0 )pk or from the corresponding (a, b, 0) distribution as pM M
k = (1 − p0 )pk /(1 − p0 ). The same
recursion used for the zero-truncated subclass applies.
The mean is 1 − pM 0 times the mean for the corresponding zero-truncated distribution. The variance is
1 − pM M M
0 times the zero-truncated variance plus p0 (1 − p0 ) times the square of the zero-truncated mean. The
probability generating function is P (z) = p0 + (1 − pM
M M
0 )P (z), where P (z) is the probability generating
function for the corresponding zero-truncated distribution.
The maximum likelihood estimator of pM 0 is always the sample relative frequency at 0.