C Formula Sheet

Probability

( )
( ) '( )
( ) ( )
( ) 1 ( )
( ) 1 ( )
( )
( ) ln( ( ))
( ) '( )
( ) ( )
H x
f x F x
F x f x
F x S x
S x F x
S x e
H x S x
h x H x
H x h x

}
}


( ) '( )
( )
( )
( )
f x S x
f x
h x
S x
=
=


( )
-
'
'
1
' 2
2 2
' ' 3
3 3 2
' ' ' 2 4
4 4 3 2
2
2
( ) or ( )
(nth Raw Moment)
(nth Central Moment)
is the mean
is the variance
3 2
4 6 3
( )
Standard Devi
k k k
x
n
n
n
n
E X x p x x f x dx
E X
E X
Var X

( =

( =

(
=

=
=
= +
= +
= =

}
3
1
3
4
2
4
ation =
Skewness is 0 if X is symmetric Skew( )=Skew( )
Kurtosis is
Coefficient of Variation is
X cX
o

o
o

= =
=


!
Combinations :
!( )!
n
n
k k n k
| |
=
|

\ .


Percentiles


Conditional Probability





Scaling
( )
( )
( ) ( )
Y X
To scale a lognormal:
X ~ lognormal ,
~ lognormal ln ,
Let
Then F ( ) Pr Pr Pr F
cX c
Y cX
y y
y Y y cX y X
c c
o
o +
=
| | | |
= s = s = s =
| |
\ . \ .

Variance
| | | | | |
| | | | | |
| |
| | ( )
( )
( ) ( )
( )
( ) ( ) ( )
2
2 2
2 2
if Independent
( )
( ) 2 , ( )
( ) ( )
1
( , ) ( ) ( )
, ( , ) ( , ) ( , ) ( , )
( , ) ( )
E XY E X E Y
E aX bY aE X bE Y
Var aX b a Var X
Var aX bY a Var X abCov X Y b Var Y
Var X E X E X
VaR X VaR X
Cov X Y E XY E X E Y
Cov A B C D Cov A C Cov A D Cov B C Cov B D
Cov A A Var A
Co
=
+ = +
+ =
+ = + +
=
=
=
+ + = + + +
=
( )
2
( , ) 0 if Independent
( , )
(Correlation Coefficient)
Bernoulli Shortcut: For any RV with only 2 values a and b:
( ) -
XY
X Y
v A B
Cov X Y
Var X a b pq

o o
=
=
=

th
p
p
p
A 100p percentile of a random variable X is a number satisfying:
Pr( )
Pr( )
X p
X p
t
t
t
s >
< s
Pr( )
Pr( | )
Pr( )
Pr( | ) Pr( )
Pr( | )
Pr( )
A B
A B
B
B A A
A B
B

=
Parametric Distributions

Distribution f(x) E(X)
( )
2
E X

Var(X)
Binomial
n trials
(1 )
k n k
n
p p
k

| |

|
\ .


np


npq
Bernoulli
1 trial
(1 )
k n k
p p


p
pq
Uniform
continuous on [d, u]
1
( )
-
x d
F x
u d u d

2
d u +

( )
2
-
if 0
3
u d
u =

( )
2
-
12
u d


Beta

1 1
( )
a b
cx x u


a
a b
u
+


( )
2
2
( 1)
ab
a b a b
u
+ + +

Exponential
(memoryless)
x
ce
u

u
2
2u
2
u

Weibull
( )
1
x
cx e
t
t
u

1
1 u
t
| |
I +
|
\ .
2
2
1 u
t
| |
I +
|
\ .


Gamma
1
x
cx e
o
u

ou

2
ou
Single Parameter
Pareto
1
c
x
o+

1
ou
o

2
2
ou
o


Double Parameter
Pareto
( )
1
c
x
o
u
+
+

1
u
o

( )( )
2
2
1 2
u
o o

| | ( )
2
2
E X
o
o

Lognormal

( )
2
2
ln
2
x
ce
x

o



2
0.5
Median
e
e
o

+
=


2
2 2
e
o +




Frailty

( )
( )
( )
( )
0
( )
Let | ( )
If a(x) is a constant then the hazard rate is exponential, otherwise Weibull
can be Gamma or Inverse Gaussian
( ) ( )
| ( )
|
( ) ( ) (Use MGF for )
Exponential a(
x
A x
h x a x
A x a t dt
H x A x
S x e
S x M A x
A
A
A = A
A
=
A = A
A =
= A
}
x) leads to a Pareto distribution
Weibull a(x) leads to a Burr distribution


Conditional Variance

( ) | | | | ( )
| | | | | | ( ) | |
| | | | | |
2 2
( ) | |
| |
|
|
i i
Var X E Var X I Var E X I
E X E E X I P A E X A
E X E E X I
Var X E Var X I
= +
= =
( ( ( =

=

Splices

1) Sum of functions must integrate to 1
2) To be continuous, functions must be equal at break point


Shifting

( )
( ) 3 1
If f 3
The mean = 1/3 + 1 (the mean of the unshifted exponential plus the shift)
e

=








Policy Limits

( )
( )
( )
( )
1
0 0
0 0
0
0
0
^ is the LIMITED EXPECTED VALUE
^ (Cost to Customer)
Definition: ( ) ( )
( ) ( ) ( )
( ^ ) ( ) ( )
( )
^ ( ) ( )
k k k
d
d
d
k
k k
k
X d
X X u
X d
d X u
E X x f x dx kx S x dx
E X x f x dx S x dx
E X d xf x dx d S d
S x dx
E X d x f x dx d S d
kx

<

=

>

( = =

= =
= +
=
= +
=
} }
} }
}
}
}
( ) ( )
1
0
( )
If (1 ) then ^ 1 ^
1
d
S x dx
d
Y r X E Y d r E X
r
(
= + = +
(
+

}


Deductibles

Ordinary Deductible of d pays max(0,X-d)
For d = 500, Loss <= 500 pays nothing, Loss of 700 pays 200

Franchise Deductible of d pays nothing if Loss is less than d, and full amount if Loss > d

Payment per Loss with Deductible

( )
( )
( ) ( )
Ordinary Deductible
0
Payment from Ins. Co.
( ) ( ) ( )
( ) ( ) ^
d d
k k
k k
d
X d
E X d
X d X d
E X d x d f x dx S x dx
E X d x d f x dx E X E X d
+

+

+
s

(
= =

>

(
= =

( (
(
= =


} }
}



Payment per Payment with Deductible

( )
| |
( )
| | | |
( )
Ordinary Deductible
|
( ) ( )
( )
1 ( )
( )
( )
( )
( ) |
^
( )
( ) ( )
( ) ( )
( )
( ) ( )
( ) Mean Excess Loss
P
P
P
X X
Y
X
X
Y
X
P
d d
Y X d X d
F x d F d
F x
F d
S x d
S x
S d
e d E X d X d
E X d
E X E X d
e d E Y
S d S d
x d f x dx S x dx
e d
S d S d
e d
+
+

= >
+
=

+
=
= >
(

(
= = =

= =
=
} }



| | | | ( )
| | | | ( ) | | ( )
( ) ( ) ( )
pmt from customer
pmt from ins. co.
^
^ ^ | Pr ^ | Pr
Average loss < d Pr Pr
E X E X d E X d
E X d E X d X d X d E X d X d X d
X d d X d
+
( = +

= < < + > >
= < + >



Franchise Deductibles

( )
Expected Payment per Loss ( )
Expected Payment per Payment ( )
E X d dS d
e d d
+
(
= +

= +









Special Cases for e(d)

Distribution e(d)

Exponential

u



Uniform ( |
0,u
2
d u


2 Parameter Pareto
1
d u
o
+

1 Parameter Pareto
( )
1
1
d
d
d d
d
u
o
o u
u
o

>

<

( | ( ) ( |
( ) ( ) ( )
( ) ( ) ( )
If Uniform 0, , then - | Uniform 0,
If Pareto , , then - | Pareto ,
If 1 Parameter Pareto , and , then - | Pareto ,
X X d X d d
X X d X d d
X d X d X d d
u u
o u o u
o u u o
+
+
+
~ > ~
~ > ~ +
~ > > ~



Loss Elimination Ratio

| |
| |
^
( )
( ) Expected % of loss not included in payment
E X d
LER d
E X
LER d
=
=









Special Cases for LER(d)

Distribution LER(d)

Exponential

1
d
e
u

2 Parameter Pareto
1
1
d
o
u
u

| |

|
+
\ .


1 Parameter Pareto
( )
1
1
d
o
u
o

Properties of Risk Measures

( ) ( )
( ) ( )
( ) ( ) ( )
( ) ( )

Translation Invariance:
Positive Homogeneity:
Subadditivity:
Monotinicity: if
A coherent risk measure satisfies all 4 properties
fails subadditivity
is coherent
E
p
p
X c X c
cX c X
X Y X Y
X Y X Y
VaR
TVaR




+ = +
=
+ s +
s s
| |
X is coherent


1
0.99
( ) ( ) Value-at-Risk
is the 99th percentile
p p X
VaR X F p
VaR
t

= =


( )
1
( )
1
( ) | ( ) Tail-value-at-Risk
( )
1
( )
1
( ) ( )
X
p p
F p
y
p
p X p
TVaR X E X X VaR X
xf x dx
p
VaR X dy
p
VaR X e VaR X

(
= >

=

= +
}
}



Distribution VaR(X) TVaR(X)


Normal

p
z o +

( )
2
0.5
where ( )
1
2
x
p
z
e
x
p
|
o |
t

+ =

Lognormal
p
z
e
o +

| |
( )
1
p
z
E X
p
o | | u
|
|

\ .



If given a mixture, use the survival function and solve for x










Maximum Covered Loss

| |
For deductible 'd' and maximum covered loss 'u'
0 X d
E X X d d X u
u d X u
s

= < s

>

| | | | | |
( ) ( )
Payment per Loss ^ ^
( )
u
d
E E X u E X d
S x dx
E X d E X u
+ +
=
=
( ( =

}


Policy Limit: the maximum amount the coverage will pay
- If Policy Limit = 10,000 and d = 500
- Pays 10,000 for loss of 10,500 or higher
- Pays Loss 500 for losses b/w 500 and 10,500

Maximum Covered Loss: the maximum loss amount that is covered
- If MCL = 10,000 and d = 500
- Pays 9,500 for loss of 10,000 or higher
- Pays Loss 500 for losses b/w 500 and 10,000
Coinsurance
| |
Coinsurance
Payment per Loss ^ ^
MCL ded
E E X u E X d o
| |
( (
=
|
( (

\ .

Coinsurance of 80% means that the insurance pays 80% of the costs

Inflation
| |
Payment per Loss (1 ) ^ ^
1 1
u d
E r E X E X
r r
o
| |
( (
= +
|
( (
+ +

\ .


Variance of Payment per Loss with a deductible

( ) ( )
( ) ( ) ( )
( )
( )
( )
( ) ( )
( ) ( )
2
Loss RV
payment per loss RV
0 Pr Pr
| |
0 Pr Pr 0 Pr Pr
L
L P
L L L
P P
X
Y
E Y X d E Y X d
Var Y E Var Y Case Var E Y Case
X d Var Y X d E Y X d X d
=
=
( (
= s + >

(
(
= +


| |
(
= s + > + s >
|

\ .



Bonus

( )
( )
( ) ( )
( )
| |
Pay Bonus of 50% of (500 - X) if 500
0.5 0,500 -
0.5 500 500,500 -
0.5 500 500,
0.5 500 ^ 500
250 0.5 ^ 500
X
B Max X
Max X
Min X
X
E X
s
=
=
=
=
=


Discrete Distributions

The (a,b,0) class

Distribution a Variance vs. Mean
Poisson

0 Variance = Mean

Negative Binomial
(Geometric is NB with r = 1)

0 >


Variance > Mean
Binomial

0 <

Variance < Mean


( )
Pr Geometric Distribution (memoryless)
1
n
N n
|
|
| |
> =
|
+
\ .


1
1
A sum of 'n' independent Negative Binomial random variables having the same
and parameters ,..., has a Negative Binomial distribution with parameters
and
A sum of 'n' independent Binomial
n
n
i
i
r r
r
|
|
=

1
1
random variables having the same
and parameters ,..., has a Binomial distribution with parameters
and
n
n
i
i
q
m m q
m
=

| |
( )
( )
1
2
1, 2,...
1
1
k
k
p b
a k
p k
a b
E N
a
a b
Var N
a

= + =
+
=

+
=

( 1) ( 1)
!
x
x x x n
n n
| | +
=
|
\ .


Probability Generating Functions

| |
| |
| |
( )
0
( )
( )
(0)
!
(0)
(1)
'(1)
''(1) ( 1)
'''(1) ( 1)( 2)
If given a primary and secondary pgf, substitute the secondary
pgf for 'z' in the primary pgf.
n
n
n
n
P
p
n
p P
P
P E X
P E X X
P E X X X

=
=
=
=
=
=



















The (a,b,1) class

1
2,3, 4,...
k
k
p b
a k
p k

= + =


Zero-Truncated Distributions

0
0
0
1
T
T
k
k
p
p
p
p
=
=


Zero-Modified Distributions

( )
( )
| | | |
( )
| |
| |
( )
0
0
0
0
0
0
0
2
0
0
1
1
1
1
1
1
( ) 1
1
is the mean of the corresponding zero-truncated distribution
is the va
M
M M
k
k
M M T
k k
M
M
M
p
p
p p
p
p p p
p
E M E Orig p E Zero Truncated
p
E N cm
Var N c c m cv
c p
m
v
>
=

= =

=
= +
=
riance of the corresponding zero-truncated distribution

Sibuya
1
ETNB with 1 0 and take lim as
1
1
T
r
a
b r
p r
| < <
=
=
=


Poisson/Gamma

( )
( )
( )
2
N ~ Poisson
~ Gamma ,
Negative Binomial ( ) Gamma( )
Negative Binomial ( ) Gamma( )
Gamma
Mean
Variance
Negative Binomial
Mean
Variance 1
Negati
The Negative Binomial is a Gamma mixture of Poissons
r
r
r

o u
o
| u
ou
ou
|
| |
=
=
=
=
=
= +
( )
( )
( ) ( ) | |
ve Binomial ( 1) is Geometric
Gamma 1 is Exponential
Weibull 1 is Exponential
where ~ Gamma
r
Var Var X E X
o
t

=
=
=
=


Coverage Modifications

Frequency Model Original Parameters Exposure Modification Coverage Modification

( )
1
Exposure
Pr 0 1
n
X > =

( )
2
Exposure
Pr 0 1
n
X > =

( )
1
Exposure
Pr 0
n
X v > =


Poisson

2
1
n
n

| |
|
\ .


v


Binomial

, m q

2
1
,
n
m q
n
| |
|
\ .


, m vq


Negative Binomial


, r |

2
1
,
n
r
n
|
| |
|
\ .


, r v|


(a,b,0) and (a,b,1) adjustments

( )
*
*
0
0 0
0
1
1 1 where * indicates revised parameters
1
M M
p
p p
p
| |

=
|

\ .


Aggregate Loss Models

Compound Variance

S=aggregate losses
N=frequency RV
X=severity RV


| | | | | |
( ) | | ( ) ( ) | |
( )
2
2
Var S
Can only be used when N and X are independent
Var S if primary distribution (# claims) is Poisson
Collective Risk vs Individual Risk
E S E N E X
E N Var X Var N E X
E X
=
= +
(
=



Convolution Method


1 2
1
...
Pr( ) ( )
Pr( ) ( )
Pr( ) ( )
( )
When given severity distributions with Pr( 0) 0
1) Modify the frequency to eliminate 0
2) Adjust the severity probabi
k
k
n N
n X
n S
S n i i i
n x i i n
p N n f n
f X n f n
g S n f n
F x g f f f
X
s + + =
= = =
= = =
= = =
=
= =

lities after removing 0









Aggregate Deductibles


| |
( ) | | | |
| |
| |
0
Assume severity is discrete
d = stop-loss or reinsurance deductible
Assume Premium ^
^ Net stop-loss Premium
Method 1 - Definition of ^
^ Pr
1 (the sum o
u
hj
j
E S d
E S d E S E S d
E S d
E S d hjg d S d
d
u
h
+
=
=
(
= =

= + >
(
=
(

| | ( )
| |
1
0
f all multiples of less than )
Method 2 - Integrate the Survival Function
^ ( ) ( )
1 (the sum of all multiples of less than )
To find ^ 2.8 where x can be 2, 4 6 or 8
u
j
h d
E S d hS hj d hu S hu
d
u h d
h
E S

=
= +
(
=
(

| | ( ) ( ) ( )
| | ( ) ( ) ( )
(0) (2) 1 (0) (2)
(0) (2) 1 (0) (2)
...
Method 1
^ 2.8 0 0 2 2 2 2.8
^ 4 0 0 2 2 4 4
amt amt d
g g g g
amt d amt
g g g g
E S P S P S P S
E S P S P S P S


| | | | | |
| | |
= = + = + >
| | |
| | |
\ . \ . \ .
| | | | |
| |
= = + = + >
| |
| |
\ . \ . \
| | ( ) ( )
dist between values of x dist b/w highest value of x below d and d
Method 2
^ 2.8 2 Pr 0 0.8 Pr 2 E S S S
|
|
|
|
.
| |
| |
= > + > |
|
|
\ .
\ .





Aggregate Coverage Modifications

If there is a per-policy deductible and you want Aggregate Payments

1) Expected Payment per Loss x Expected Number of Losses per Year

| | | | ( ) E S E N E X d
+
( =


OR

2) Expected Payment per Payment x Expected Number of Payments per Year

| | | |
( )
'
'
|
where N is the number of positive payments frequency Pr( )
E S E N E X d X d
X d
(
= >

>


1) Better for discrete severity distributions

2) Better for if severity is Exponential, Pareto or Uniform

Exact Calculation of Aggregate Loss Distribution

Two cases for which the sum of independent random variables has a simple distribution
1) Normal Distribution. If i
X

are Normal with mean

and variance
2
o
, their sum is normal.

2) Exponential or Gamma Distribution. If i
X

are exponential or gamma, their sum has a gamma
distribution




















Normal Distributions

( )
2
i
2
2
If n random variables X are independent and normally distributed with parameters and ,
their sum is normally distributed with parameters and .
Calculate Pr | 1 using and
Calculate Pr
n n
S c n
S
o
o
o > =
> ( )
2
1 2,
| 2 using 2 and 2 , etc.
Then multiply each these probabilities by their respective , etc.
c n
p p
o =



Exponential and Gamma Distributions

The sum of exponential random variables with common mean
is a Gamma distribution with parameters and . When a gamma
distribution's parameter is an integer, the gamma distribution is called
an
n
n
u
o u
o
=
Erlang distribution.
In a Poisson Process with parameter , the number of events occurring by time t
has a Poisson distribution with mean .
In a Poisson Process with parameter , the tim
t

e between events follows

1
an exponential distribution with parameter .
1
In a Poisson Process with paremeter , the time between events is exponential
with mean . Therefore, the time until the nth

u
u
( )
( )
1
0
-
-
event occurs is Erlang with parameters and
The probability of exactly events occurring before time is
!
Gamma CDF
( ) 1
!
If 1, ( ) 1-
If 2, ( ) 1-
x
n
x
j
n
X
j
x
x
n
x
e
n x
n
x
e
F x
j
F x e
F x e
u
u
u
u
u
u
u
o
o

=
| |
|
\ .
| |

\ .
=
= =
= =

-
x
x
e
u
u
| |
| |
\ .
| |

|
\ .


Empirical Models

( )
( )
Bias

is the estimator, is the parameter being estimated

bias is the expected value of the estimator minus its true value
Estimator is unbiased if 0 for all
Estimator is asympto
bias E
bias
u
u
u u u u u
u u
(
=

=
( )
( )
tically unbiased if lim 0
The sample variance (with division by -1) is an unbiased estimator of variance
Sample mean is an unbiased estimator of the true mean
Consistency (weak consistency)
Defi
x
bias
n
u
u

=
( )
( )
( )
nition: Estimator is consistent if lim Pr 1 for all 0
1) An estimator is consistent if it is asymptotically unbiased
and Var 0 as n
2) The MLE is always consistent
3) If 0 then is
n
n
n
MSE
u
u u o o
u
u u

< = >

( )
( )
( )
( )
( )
( )
2
2
consistent
Mean Square Error
|
MSE is a function of the true value of the parameter
MSE E
MSE Var bias
u
u u
u u u u
u u u
(
=
(

= +





Complete Data

Grouped Data

( )
)
1
1
2
( )
where x is in ,
= # points in the interval
= total points
( ) Histogram
( ) Ogive
To find the 2nd raw moment, calculate ( )
If there is a policy limit (say 8000)
j
n
j j
j j
j
n
n
b
n
a
n
f x
n c c
c c
n
n
f x
F x
f x x dx

=
=

}
( )
2
10,000 8000
2 2
5000 8000
, then E X^8000 would have as its last 2 terms:
( ) ( ) 8000
n n
f x x dx f x dx
(

+
} }

Variance of Empirical Estimators

( )
Binomial:
Variance = (1 )
X
If Y= (Binomial Proportion),
m
(1 )
Variance =
m
Multinomial:
Variance (1 ) i = category
Covariance -
X (1 )
If Y= , Variance =
m m
-
Covariance
i i
i j
i i
i
mq q
q q
mq q
mq q
q q
q

=
=

=
j
q
m



Individual Data


( ) ( )
( ) ( ) ( )
( ) ( )
( ) ( ) ( )
( ) ( )
( )
( ) ( )
( )
3
3
1
if S is known
1
where is the # of survivors past time x
| |
The empirical estimators of ( ) and ( ) are unbiased
n
n n
n
x x
n x
x y y
x x y x x y x x
x
S x S x
Var S x
n
S x S x
Var S x
n
n n n
Var S x n
n
n n n
Var p n Var q n
n
S x f x

= =



Individual Data

1) Determine the estimator
2) Determine what's random and what's not
3) Write an expression for the estimator with symbols for random variables
4) Calculate the variance of the random variables
5) Calculate the variance for the whole expression
(i.e. ( ), ( ), etc.) Var aX bY Var aX +














Kaplan Meier Product Limit

( )
( )
( )
0
1
1
0 0
-
( ) 1
( ) where t is the end of the study Exponential Extrapolation
risk set
death
entry time
withdrawal time
death time
Pr
1 1 2
Shortcut:
1 0.5
j
i
n
i
i
t
t
i
i
i
i
i
s
S t
r
S t S t
r
s
d
u
x
S x X x
n n n

=
| |
=
|
\ .
=
=
=
=
=
=
= >
+ ~

[

Nelson Aalen Estimator

1
1

( )

( )

( )
j
i
i
i
H x
s
H t
r
S x e

=
=

Lives that leave at the same time as a death are in the risk set.
Lives that arrive at the same time as a death are not in the risk set.
Censored lives are in the risk set but are not counted as deaths

k = # distinct data points

Confidence Intervals

For S(x), the boundaries must be between 0 and 1.
For H(x), the boundaries can be anything.






Calculator Shortcuts

1) Enter in column 1
2) Enter the formula ln(1- 1) for column 2
3) Select L1 as first variable and L2 as second
4) Calculate -
i
i
x y
Nelson Aalen Kaplan Meier
s
r
L
e e

Estimation of Related Quantities

If using Kaplan-Meier or Nelson-Aalen methods, E(X^d) = area under the curve of S(x). Multiply
the base x height of each of the rectangles.

Bayes Theorem

( )
( ) ( )
( ) ( ) ( ) ( ) ( ) ( )
1 1
1
1 1 2 2
|
|
| | ... |
n n
P E A P A
P A E
P E A P A P E A P A P E A P A

=
+ + +

( )
( )
( )
( ) ( ) ( )
( )
2
2
Greenwood's Approximation of Variance (Kaplan-Meier)
( ) ( )
1
( ) if data is complete (no censoring or truncation)
Variance of Nelson-Aalen Estimator
( )
Lin
j
j
j
j j j
y t
n n
j
j
y t
s
Var S t S t
r r s
S x S x
Var S t
n
s
Var H t
r
s
s
=

=
=

( ) ( )
( )
( )
0.5(1 ) 0.5(1 )
1
0.5(1 )
ear Confidence Intervals
( ) ( ) , ( ) ( )
Log-transformed confidence interval for S(t)
( )
( ) , ( ) where exp
( ) ln ( )

L
n p n n p n
p n
U
U
n n
n n
S t z Var S t S t z Var S t
z Var S t
S t S t U
S t S t
+ +
+
+
| |
| |
|
=
|
|
\ .
\ .
( ) 0.5(1 )
og-transformed confidence interval for H(t)
( )
( )
, ( ) where exp
( )
p
z Var H t
H t
H t U U
U
H t
+
| |
| |
|
=
|
|
|
\ .
|
\ .



Kernel Smoothing

( )
( )
( )
1
1
Uniform
1
( )
2
0
0
( )
2
1
1

( ) ( )
1

( ) ( )
is a sample point
is the
i
i
i
n i
i
n i
i i
x
i
i
x i i
i
n
x
i
f x probability
n
x
i
f x probability
i
x b x x b
k x
b
Otherwise
x x b
x x b
K x x b x x b
b
x x b
f x k x
n
F x K x
n
x
x
=
=
=
=

s s +

= s s +

> +

| |
=
|
\ .
| |
=
|
\ .

6 10
estimation point
Kernel distribution is 1 for observation points more than one bandwidth to the left
Kernel distribution is 0 for observation points more than one bandwidth to the right
K (13) K (13) K > >
25
12 12 12
(13)
ex) To find K (11), linearly interpolate between K (7) and K (17)
Triangular
1
Height of triangle is
Base of triangle is 2
b
b



| |
| | | |
2
2
Expected Values
| ( is the original random variable)
The mean of the smoothed distribution is the same as the original mean
Uniform Kernel
( ) ( )
3
Triangular Kernel
( ) ( )
6
E X Y Y Y
E X E Y
b
Var X Var Y
b
Var X Var Y
=
=
= +
= +

































Approximations for Large Data Sets

)
(
(
(
1
1
1
1
'
is the # of left truncated observations in ,
- number of new entrants
is the # of right censored observations in ,
is the # of losses in ,
is the risk set ,

j j j
j j j
j j j
j j j
j
d c c
u c c
x c c
r c c
q
+
+
+
+

(
( )
1
'
1
is the decrement rate in ,
# withdrawals
# survivors
All entries/withdrawals at endpoints
All entries/withdrawals uniformly distributed
0.5
j j
j
j
j
j j j j j
j
j
j j j
j j j
j j j j
c c
x
q
r
P P d u x
v
w
v w u
r P d
r P d v
+
+
(

=
= +
=
=
+ =
= +
= +

Multiple Decrements

( ) (1) (2)
'( )
1 2
3
1 2
'( ) '( )
3 3
' ' ...
1 1 1
1
x
t t t
t t
t t t
x x
t t t t
p p p
x x x
p
r r r
q p
t
+ +
+
+ +
+ +
=
| || || |
=
| | |
\ .\ .\ .
=



Parametric Models

Method of Moments

( )
2
1 1
n n
i i
i i
x x
m t
n n
= =
= =




Distribution Formulas Formulas
Exponential

m u =


Gamma

2
2

m
t m
o =

t m
m
u

=


Pareto

2
2
2 2

2
t m
t m
o

=

2
mt
t m
u =

Lognormal

2ln( ) 0.5ln( ) m t =

2
2ln( ) ln( ) m t o = +

Uniform on | |
0,u

2m u =


When they dont specify which moment to use, use the first k moments, where k is the number
of parameters youre fitting.

For an inverse exponential, add the reciprocals to get the mean.


Percentile Matching

( 1)

if ( 1) is an integer
Otherwise multiply ( 1) and interpolate
The smoothed empirical percentile is not defined if the product is less than 1 or greater than
p n p
x n p
n p
n
t
+
= +
+




Maximum Likelihood

Type of Data Formula
Discrete distribution, individual data

x
p

Continuous distribution, individual data

( ) f x

Grouped Data

( ) ( )
1 j j
F c F c

Individual Data censored from above at u

1 ( ) for censored observations F u

Individual Data censored from below at d

( )for censored observations F d

Individual Data truncated from above at u

( )
( )
f x
F u

Individual Data truncated from below at d

( )
1 ( )
f x
F d


Cases where MLE = Method of Moments Estimator
(if no censored or truncated data)

Distribution Result
Exponential MLE = MoM
Gamma MLE = MoM if fixed
o

Normal MLE = MoM
Poisson MLE = MoM
Negative Binomial MLE = MoM if r is known
Binomial MLE = MoM if m is known


If the MLE is the sample mean, the variance of the MLE is the variance of the
distribution =
( ) Var X
n


Common Likelihood Functions

MLE Function MLE
( )
b
a
L e
u
u u

=

b
a
u =

( )
a b
L e
u
u u

=

a
b
u =

( ) ( )
1
b
a
L u u u =

a
a b
u =
+

MLE Formulas

Distribution Formula CT?


Exponential
( )
1

n c
i i
i
x d
n
u
+
=

=

Yes





Lognormal
1
ln

n
i
i
x
n

=
=

( )
2
2
1
ln

n
i
i
x
n
o
=
=

No
Inverse Exponential
1

1
n
i
i
n
x
u
=
=

No


Weibull, fixed
t

1 1

n c n c
i i
i i
x d
n
t t
t
u
+ +
= =

=




Yes
Uniform on Individual Data( |
0,u
{ }
max
i
x u =

No

Uniform on Grouped Data( |
0,u

OR Some observations are censored
at a single point
There must be at least one observation
above
j
c

Upper bound of highest finite interval

Number of observations below
j
j
j
j j
n
c
n
c
n c
u
| |
=
|
|
\ .
=
=



No

Uniform on Grouped Data( |
0,u

All groups are bounded

Min of
1) UB of highest interval with data
2) LB of highest interval with data *
j
n
n
u =
| |
|
|
\ .




n = # of uncensored observations
c = # of censored observations
d = truncation point
x = observation if uncensored or the censoring point if censored
CT = formula can be used for left-truncated or right-censored data

Bernoulli Technique

Whenever there is one parameter and only 2 classes of observations, maximum likelihood will
assign each class the observed frequency, and you can then solve for the parameter.

If X can be only 2 values (a or b)
( )
# data points = a
# data points
P X a = =









Two-parameter Pareto, fixed
u

( ) ( )
1 1

ln ln
n c n c
i i
i i
n
K
K d x
o
u u
+ +
= =
=
= + +




Yes
One-parameter Pareto, fixed
u

( )
1 1

ln max , ln
n c n c
i i
i i
n
K
K d x
o
u
+ +
= =
=
=




Yes


Beta, fixed
u

b = 1

1

ln ln
n
i
i
n
a
K
K x n u
=
=
=

No


Beta, fixed
u

a = 1

( )
1

ln ln
n
i
i
n
b
K
K x n u u
=
=
=

No
Reasons to Use Maximum Likelihood

1) Method of Moments and Percentile Matching only use a limited number of features from the
sample.

2) Method of moments are hard to use with combined data.

3) Method of moments and percentile matching cannot always handle truncation and censoring.

4) Method of moments and percentile matching require arbitrary decisions on which moments
or percentiles to use.

Reasons NOT to Use Maximum Likelihood

1) Theres no guarantee that the likelihood can be maximized it can go to infinity.

2) There may be more than one maximum.

3) There may be local maxima in addition to the global maximum; these must be avoided.

4) It may not be possible to find the maximum by setting the partial derivatives to zero; a
numerical algorithm may be necessary.
























Fishers Information

1 Parameter

( ) ( )
( )
( )
2
2
Fisher's Information
1
Var
d
E l
d
u u
u
u
u
(
I =
(
(

=
I


2 Parameters

( )
( ) ( )
( ) ( )
( )
( ) ( )
( ) ( )
( )
2
2
2
2
1
, ,
,
, ,

Var ,
l l
I E
l l
Var Cov
Cov Var
o u o u
o u
o
o u
o u o u
o u
u
o ou
o u ou
ou u

| |
c c c
|
c c
c
|
=
|
c c c
|
|
c c
c \ .
| |
= = I
|
|
\ .

Inverting a Matrix

1
1
a b d b
c d c a ad bc

| | | |
=
| |

\ . \ .


Delta Method

( )
( )
( ) ( ) ( )( )
2
2
2
1
g( ) ( ) 1 Variable
g( , ) ( ) 2 ( , ) ( ) 2 Variables
g( ) General
where ,..., and is the covariance matrix
Take
k
dg
Var X Var X
dx
g g g g
Var X Y Var X Cov X Y Var Y
x x y y
Var X g g
g g
g
x x
| |
~
|
\ .
| | c c c c
| |
~ + +
| |
c c c c
\ .
\ .
'
~ c E c
'
| | c c
c = E
|
c c
\ .
derivative with respect to unknown variable



Fitting Discrete Distributions

Distribution Method of Moments MLE
Poisson

x =

x =



Negative Binomial
2 2
2
2

Sample Variance (divide by n)

x x
r
x x
o
|
o
o

= =

r x | =


Binomial

x
q
m
=


Choosing between (a,b,0) distributions to fit the data:

2
-1
1) Compare sample variance to sample mean x
Poission: Variance = Mean
Negative Binomial: Variance > Mean
Binomial: Variance < Mean
2) Calculate and observe the slope as a function of k
If ratio
k
k
kn
n
o
s are increasing, then a > 0
Poisson = 0 slope
Negative Binomial = positive slope
Binomial = negative slope
n is the # of policies/observations of k


The variance of a mixture is always at least as large as the weighted average of the
variances of the components and usually greater due to:

| | | | ( )
variance of a mixture
weighted average of variances
( ) ( | ) | Var X E Var X I Var E X I = +




Asymptotic Variance of MLEs

Distribution Formula
Exponential
( )
2

Var
n
u
u =
Uniform ( |
0,u

( )
( )
2
2

1 ( 2)
n
Var
n n
u
u =
+ +

Weibull fixed
t

( )
2
2

Var
n
u
u
t
=

Pareto fixed
u

( )
2
Var
n
o
o =

Pareto fixed
o

( )
( )
2
2

Var
n
o
u
u
o
+
=


Lognormal
( )
2
Var
n
o
=

( ) , 0 Cov o =

( )
2

2
Var
n
o
o =


Poisson

( )

Var
n

=




( )
1
( ) Var X Var X
n
=









Hypothesis Tests Graphic Comparison

( ) ( )
*( )
1 ( )
( )
*( )
1 ( )
F x F d
F x
F d
f x
f x
F d

( ) plots
( ) ( ) *( )
Empirical - Fitted
Empirical calculation uses a denominator of n
If ( ) 0, then ( ) *( )
had more data x than predicted by model
If ( ) 0, then ( ) *( )
had less data
n
n
n
D x
D x F x F x
D x F x F x
D x F x F x
=
=
> >
s
< <
x than predicted by model
If data truncated at , ( ) 0
1
Every vertical jump has distance
d D d
n
s
=


| |
( ) ( ) ( )
plots
1
On horizontal axis, one point every multiple of
1
Domain and Range of Graph are 0,1
Points are , *
The first data point corresponds to the first sample value
If slope is less than 45 , f
n j j
p p
n
F x F x

+
itted distribution is putting
too little weight in that region
If slope is more than 45 , fitted distribution is putting
too much weight in that region
Don't plot censored values


Kolmogorov-Smirnov Test

0
max ( ) *( )
Max occurs right before or after a jump
If D < critical value, do not reject H (null hypothesis)
Having censored data lowers D and also lowers the critical value
Critical values 0 a
n
d x u
D F x F x
< <
=

( )
( )
4 4
s n
If data are X = 2000, 4000, 5000, 5000
and the 5000 values are right-censored
Then F 5000 F 5000 0.5

= =




Anderson-Darling Test

( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( )
2
2
1
0
2
1
1
*( ) ln * ln *
ln * ln *
n includes censored observations
k
n j j j
j
k
n j j j
j
A nF u n S y S y S y
n F y F y F y
+
=
+
=
= +
+

Chi-Square

( )
2
1
2
1
# of observations in each group
(Expected # of observations in each group)
Total # of observations
Each group should have at least 5 expected observations
k
j j
j
j
k
j
j
j
j
j j
O E
Q
E
O
Q n
E
O
E np
n
=
=

=
| |
|
=
|
\ .
=
=
=

. If not,
you have to bring some groups together.
Degrees of Freedom
Distribution with parameters is given
1 DoF
Distribution is fitted/estimated by MLE or using different data
1 DoF Parameters are fi
k
k r

{
( )
2
1
tted from the data
# parameters
# groups
Independent Periods
where is the variance
Degrees of Freeedom - (p is number of estimated parameters)
k
j j
j
j
j
r
k
O E
Q V
V
k p
=
=
=

=
=

Kolmogorov-
Smirnov
Anderson-Darling Chi-square Loglikelihood

Individual Data


Individual Data

Individual or Grouped Data

Individual or Grouped
Data

Continuous Fits


Continuous or Discrete Fits


If there is censored
data <
, should
lower critical value


If there is censored data
<
,
should lower critical value


If there is censored data
<
, no adjustment of
critical value


If there is censored data
<
, no adjustment of
critical value


If parameters are
fitted, critical value
should be lowered


If parameters are fitted,
critical value should be
lowered


If parameters are fitted,
critical value adjusts
automatically


If parameters are fitted,
critical value adjusts
automatically

Larger sample size
makes critical value
decline


Critical value independent of
sample size

Critical value independent of
sample size

Critical value independent
of sample size

No Discretion in
grouping of data


No Discretion in grouping of
data


Discretion in grouping of data


Uniform weight on
all parts of
distribution


Higher weight on tails of
distribution

Higher weight on intervals
with low fitted probability


0
1
Type I Error: Rejecting H when it is true
Type II Error: Rejecting H when it is true














Likelihood Ratio Algorithm

If parameters are added to the model, the new model will have a loglikelihood at least as great.

The # DoF for the likelihood ratio test is the number of free parameters in the alternative model
minus the number of free parameters in the base model (null hypothesis).

1 2
1
2
1 2
Compare 2( ) to critical value at selected chi-square percentile and DOF
Alternative Model Loglikelihood (which will be higher)
Base Model
If 2( ) > critical value, accept
LogL LogL
LogL
LogL
LogL LogL

=
=

1 2
1 2
alternative hypothesis
Start by comparing best 2-parameter to best 1-parameter
If 2( ) Critical Value (it fails), compare best 3-parameter distributions to best 1-parameter
If 2( ) C
LogL LogL
LogL LogL
<
> ritical Value (it passes), compare 3-parameter distributions to best 2-parameter


Schwarz-Bayesian Criterion

- ln
2
where r is the # parameters
where n is the # of data points
The distribution with the highest resulting LogL is selected
r
LogL n
| |
=
|
\ .
















Credibility

( ) ( )
2
0
2
0
-1
p p
where
1
coefficient from the standard normal =
2
Given y , P% = 100 2 percent corresponding to y 1
maximum fluctuation you will accept (i.e. within 5%)
F
p
p
e n CV
y
n
k
p
y
k
=
| |
=
|
\ .
+
| |
= u
|
\ .

=


Limited Fluctuation Credibility: Poisson Frequency

| |
2
2
2
All must be the same!
1
s
E X
CV
E X

(

+ =



Credibility for
Experience
expressed in
Number of Claims Claim Size (Severity) Aggregate Losses/Pure
Premium

Exposure Units
F
e


0
n

( )
2
0
s
n
CV

( )
2
0
1
s
n
CV

+


Number of Claims
F
n


0
n

( )
2
0 s
n CV
( )
2
0
1
s
n CV +



Aggregate Losses
F
s

0 s
n

( )
2
0 s s
n CV
( )
2
0
1
s s
n CV +




Pure Premium is the expected aggregate loss per policyholder per time period.


Limited Fluctuation Credibility: Non-Poisson Frequency

2
0 F s
F F f
e n CV
n e
=
=



Credibility for
Experience
expressed in
Number of Claims Claim Size (Severity) Aggregate Losses/Pure Premium

Exposure Units
F
e

2
0
2
f
f
n
o

| |
|
|
\ .

2
0
2
s
s f
n
o

| |
|
|

\ .

2
2
0
2 2
f
s
f s f
n
o
o

| |
+
|
|

\ .


Number of Claims
F
n


2
0
f
f
n
o

| |
|
|
\ .

2
0
2
s
s
n
o

| |
|
\ .

2
2
0
2
f
s
f s
n
o
o

| |
+
|
|
\ .


Aggregate Losses
F
s

2
0
f
s
f
n
o

| |
|
|
\ .

2
0
2
s
s
s
n
o

| |
|
\ .

2
2
0
2
f
s
s
f s
n
o
o

| |
+
|
|
\ .


# of Insureds is Exposure

Partial Credibility

( )
( )
1
Credibility Premium
Manual Premium
Z Credibility
Observed Mean
Expected Claims
Number of Expected Claims needed for Full Credibility
C
C
C
F
F
P ZX Z M
P M Z X M
P
M
X
n
Z
n
n
n
= +
= +
=
=
=
=
=
=
=



Bayesian Credibility

Bayesian Methods Discrete Prior

Class 1 Class 2
1) Prior Probabilities
2) Likelihood of Experience
3) Joint Probabilities
Product of rows above Product of rows above
4) Posterior Probabilities
Quotients of row 3 over row 3 sum Quotients of row 3 over row 3 sum
5) Hypothetical Means
6) Bayesian Premium
Product of rows 4 and 5 Product of rows 4 and 5

Bayesian Premium is the predicted expected value of the next trial

# claims is Bernoulli means at most 1 can occur

Bayesian Methods Continuous Prior

( )
( ) ( )
( ) ( )
( ) ( ) ( )
( )
1
1
1
1 1 1 1
,..., |
| ,..., Posterior Density
,..., |
Limits of Integration are according to prior distribution
| ,..., | | ,..., Predictive Density
is the prior density
n
n
n
n n n n
f x x
x x
f x x d
f x x x f x x x d
t u u
t u
t u u u
u t u u
t u
+ +

=
}
}
( )
( ) ( )
( ) ( )
( )
( )
1
1
1
1
1
1 1
| ,..., is the posterior density
,..., | is the likelihood function of the data given conditional
,..., | |
,..., is the unconditional joint density function
,..., ,.
n
n
n
n i
i
n
n
x x
f x x
f x x f x
f x x
f x x f x
t u
u u
u u
=
=
=
[
( ) ( )
.., |
n
x d u t u u
}


1
0
!
n t
n
n
t e dt
o
o

+
=
}

Bayesian Credibility: Poisson/Gamma

( )
( )
*
*
*
*
* *
*
* *
~
1
~ ,
+ claims
exposures
1
Posterior: ,
Posterior mean is the avg. # claims/policy
Predictive: Negative Binomial ,
c
N Poisson
Gamma
P
Gamma
r

o
u
o o

o

o u

o | u
| |
=
|
\ .
=
= +
=
| |
=
|
\ .
= =

Bayesian Credibility: Normal/Normal

( )
( )
*
*
*
*
~ ,
~ ,
Observed Average
Exposure
Posterior Mean
Posterior Variance
Predictive Mean
Predictive Variance
X Normal v
Normal a
x
n
v anx
v an
va
a
v an
v a
u
u

=
=
+
=
+
=
+
=
= +

Bayesian Credibility: Lognormal/Normal

( )
( )
| |
( )
* *
*
*
0.5 0.5 0.5 0.5
,
,
ln
Find
Posterior Mean
Posterior Variance
|
i
a v v v
X Lognormal v
Normal a
x
x
n
v anx
v an
va
a
v an
E X E e E e e e e
u u
u
u

u
+ +
=
+
=
+
=
+
( ( = = =

Bayesian Credibility: Bernoulli/Beta

( )
1 1
*
*
Probability of a claim = q
0,1
Uniform is a special case of Beta distribution with 1 and 1
# claims
n = exposure
(1 )
claims
Plug into Posterior Dis
exposures - claims
a b
q Unif
a b
k
Beta Cx x
a a
b b
u

= = =
=
=
= +

`
= +
)
| |
*
* *
* *
* * * *
tribution
|
If 1, treat as a series of 'm' Bernoullis
If exposure is 2 years, treat as '2m' Bernoullis
1) 2
and
2)
( 1) ( )
a
E x
a b
m
n m
a a
m
a b a b
x x x
u =
+
>
=
| |

|
+ +
\ .
I + = I

Bayesian Credibility: Exponential/Inverse Gamma

( )
( )
( )
( )
( )
1
*
*
*
*
1
*
*
1
| Exponential
Inverse Gamma
(next loss)
1
If ( ) is Gamma instead of exponential
1
|
x
x
f x e
e
n
nx
E
f x
f x x e
n
nx
u
|
o
u
o
q
u
q
u
u
|
t u
o u
o o
| |
|
o
u
q u
o o q
| |

=
=
I
= +
= +
=

=
I
= +
= +



Loss Functions

( ) ( )
( )
2
For the loss function minimizing MSE

,
Bayesian Point Estimate is the mean of the posterior distribution
For the loss function minimizing Absolute Value of the Error

,
Bayesian Point Est
l
l
u u u u
u u u u
=
=
imate is the median of the posterior distribution
For the zero-one loss function
Bayesian Point Estimate is the mode of the posterior distribution


Buhlmann Credibility

Buhlmann Credibility: Basics

( ) ( )
( )
( )
Overall Mean Expected Value of the Hypothetical Mean
Expected Value of the Process Variances
Variance of the Hypothetical Mean
Overall Variance
For Poisson frequency HM
E
v E v
a Var
a v

O
O
O
= O (

= O (

= O (

+ =
= PV
Buhlmann's k

Buhlmann's Credibility Z
Z
# periods when studying frequency or aggregate losses
# claims when studying severity
If given 2 classes and there are multipl
v
k
a
n na
n k na v
n
n
=
= =
+ +
=
=
e groups within each class,
you must find the mean and variance of each group separately.

Buhlmann Credibility: Continuous

Given a distribution function and the prior function:
1) Use the distribution function to get the HM and PV
2) Find v and a using the prior distribution
Model Prior Posterior Predictive Buhlmann v Buhlmann a
Poisson
( )

Gamma
1
o

u
=

Gamma
*
*
claims
= exposures
o o

= +
+

Negative Binomial
*
*
*
1
=
r o
| u

=
=



ou



2
ou

Bernoulli

(q)

Beta

a
b

Beta

*
*
claims
= exposures - claims
a a
b b
= +
+
Bernoulli
| |
*
* *
|
a
q E x
a b
u = =
+

( )( 1)
ab
a b a b + + +

2
( ) ( 1)
ab
a b a b + + +
Normal

( ) , v u

Normal

a

Normal

*
*
v anx
v an
va
a
v an

+
=
+
=
+

Normal


*
2
*
a v

o
=
= +


v


a
Exponential

( ) u


Inverse Gamma

o
u

Inverse Gamma

*
*
n
nx
o o
u u
= +
= +

Pareto

*
*
o o
u u
=
=


2
( 1)( 2)
u
o o

2
2
( 1) ( 2)
u
o o

Exact Credibility

If you have conjugate pairs and they ask for a Buhlmann estimate, use the Bayesian estimate.


Buhlmann as Least Squares Estimates of Bayes

| | | |
( )
| |
| |
( ) | | | |
2
2
E Initial probabilities x Outcomes = E Initial probabilities x Bayesian Estimates

,
( )
(1 )
( )
,
where X are the initial outcomes and Y are the Baye
i i
i i
i i i
Y X
Cov X Y
Z
Var X
Z E X
Var X p X E X
Cov X Y p X Y E X E Y
o |
|
o
= +
= =
=
=
=

| |
| |
| |
first observation = 0
sian observations
Buhlmann Predictions
( 0 ) (1 )
(2) (1 ) 2
(8) (1 ) 8
Graphics Questions
1) The Bayesian prediction must be within the range of the hypothetical means
-w
c
c
c
P Z E X
P Z E X Z
P Z E X Z
=
= +
= +
ithin range of the prior distribution
2) The Buhlmann predictions must lie on a straight line
3) There should be Bayesian predictions both above and below the Buhlmann line
4) The Buhlmann prediction must be between the overall mean and the observation


( )
( )
( )
i j
i j
i
Cov X X
Cov X X a
Var X v a
=
= +



Empirical Bayes Non-Parametric Methods

Uniform Exposures Non-Uniform Exposures


Mean of all data

x

x



v
Mean of sample
variances of the
rows

2
1 1
avg/class
avg/cell
1
( )
( 1)
r n
ij i
i j
x x
r n
= =

2
1 1
avg/class
avg/cell
1
per class
( )
( 1)
r n
ij ij i
i j
r
i
i
m x x
n
= =
=

a


2
overall avg 1
avg/class
1
( )
1
r
i
i
v
x x
r n
=
| |
|

|
\ .

2
overall avg 1
exp/class avg/class
2
1
exp/class
overall
overall
( ) ( 1)
r
i i
i
r
i
i
m x x v r
m
m
m
=
=

n


Years of Experience




# Policyholders


r = # groups
m = # exposures


To calculate individual variances

( )
2
1
1
i
X X
v
n

=

C
In the P formula, M = the average of ALL claims


Empirical Bayes Semi-Parametric Methods

( )
2
2
2
Poisson Model

1
# Policyholders

regardless of # of years (but if non-uniform exposures, use n = # exposures

for the group you are looking at)
If non-uniform exposures
i
x
v x
a s v
X X
s
r
r
a
Z
a v
=
=
=

=
=
+

C
C
, must be calculated using Non-parametric formula
For P
1) X= total # observed claims (but if non-uniform exposures, use the average)
2) If exposure is 5 years, divide P by 5 to get estimate for 1 y
a
| |
( )
| |
( )
2
2
2
ear (next year)
Non-Poisson Model
1) Negative Binomial with fixed
|
| (1 )

(1 )

2) Gamma with fixed
|
|

E N r r
Var N r r
x
v x
a s v
E X
Var X
x
v x
a s v
|
|
| |

|
u
o ou
o ou

u
=
= +
=
= +
=
=
=
=
=
=

Simulation

Inversion Method

| )
1 2 2
1) Get ( )
2) Solve for x
3) Plug in 'u' to get simulated value
If (2 ) .25
(2) .75
Then .25 .75 is mapped to 2
If F(x) = 'a' (constant) in range , , then map a x
If given a graph with (x, F(x))

u F x
F
F
u x
x x

=
=
=
s s =

1) Start on the y-axis with the u values

2) Move right until you hit the line
If the line is horizontal, keep going right until it starts going up
3) Go vertically down to x
4) x is the simulated value















Number of Data Values to Generate

( )
( )
2
0
( ) 1 ( )

( )
F
F x F x
Var F x
n
e n CV

=
=


Estimated Item

Mean

Confidence Interval:
n
s
x z
n
t
| |

|
\ .


n
s is the square root of the unbiased sample variance after n runs

Number of Runs:

Calculates number of runs needed for the sample mean to be within 100k% of the true mean.

2 2
0
Must use unbiased Variance in n n CV CV >


Remember that
( ) ( )
1
Var X Var X
n
=


F(x)

Confidence Interval:


( )
( )
( ) ( )
( ) 1 ( )
( )
F x z VaR F x
F x F x
F x z
n
t
t

Number of Runs:


0 0
1
# runs below x
n
n
n
n
n
P
n P
n
n n n
P
P
n
P
| |

| | |
> =
| |
\ .
|
\ .
=

Percentiles
q
t


Confidence Interval:


| |
( )
( )
1
2
1
2
,
0.5 1
0.5 1
a b
p
p
Y Y
a nq z nq q
b nq z nq q
+
+
(
= +
(

(
= + +
(
(

































Risk Measures

( )
| |
1
( )
1
( ) |
^
( ) ( )
1
( )
1
( )
1
( ) is the mean of the upper tail of the distribution
( ) Conditional Tail Expectation
( )
X
p p
p
p X p p
F p
y
p
q
q
q
TVaR X E L L VaR
E L E L
VaR X e VaR X
p
xf x dx
p
VaR X dy
p
TVaR X
TVaR X
Y
TVaR X
t
t

( = >

(

= + = +

=
=
}
}
( )
( )
( )
( )
2 2
2
2
2
1
( ) ( )
1
( ) ( )
( )
1
Confidence Interval = ( ) ( )
n
j
j k
q q
q
q q
q
q
q q
n k
n
s E TVaR X E TVaR X
n
s q TVaR X VaR X
Var TVaR X
n k
TVaR X z Var TVaR X
t
=
+
| |
(
(
=
|
(

\ .
+
=
+

951
Estimate of ( ) is where 1
So if simulation has 1000 runs and you're estimating 95th percentile,
Then use
q k
VaR X Y k nq
Y
= +
(



Bootstrap Approximation

( )
( ) ( ) ( ) ( )
( )
( )
( )
1
1
2
,...,
1
2
2
1
2
2
1
2
( ) is the parameter
,..., is an estimator based on a sample of 'n' items
( ) ,..., ( )
Estimating the mean with the Sample Mean

n
n
n
g x x
F n n
n
i
i
n
i
i
x
F
g x x
MSE F E g x x F
x x
n
x x
MSE
n n
u
u u
o
o

=
=
(
=

= =

Sums of Distributions

Single Multiple
Bernoulli Binomial
Binomial Binomial
Poisson Poisson
Geometric Negative Binomial
Negative Binomial Negative Binomial
Normal Normal
Exponential Gamma
Gamma Gammas
Chi-Square Chi-Square