Antitrust Notice

1. The Casualty Actuarial Society is committed to adhering strictly to the letter and
spirit of the antitrust laws Seminars conducted under the auspices of the CAS spirit of the antitrust laws. Seminars conducted under the auspices of the CAS
are designed solely to provide a forum for the expression of various points of
view on topics described in the programs or agendas for such meetings.
2. Under no circumstances shall CAS seminars be used as a means for competing
companies or firms to reach any understanding expressed or implied that companies or firms to reach any understanding expressed or implied that
restricts competition or in any way impairs the ability of members to exercise
independent business judgment regarding matters affecting competition.
3. It is the responsibility of all seminar participants to be aware of antitrust
regulations to prevent any written or verbal discussions that appear to violate regulations, to prevent any written or verbal discussions that appear to violate
these laws, and to adhere in every respect to the CAS antitrust compliance
policy.
CREDIBILITY FOR A TOWER OF
EXCESS LAYERS EXCESS LAYERS
Munich Reinsurance America, Inc.
Dave Clark
Agenda
1. Basic Credibility Concepts
2. Multivariate Credibility
3. The XOL Reinsurance Problem
4. The Recursive Form for XOL Pricing
Preliminaries
Criteria for an estimator of future losses:
U bi d th t d l f th ti t i l t th t t d Unbiased = the expected value of the estimator is equal to the true expected
loss

Minimum Variance = on average the value produced by this estimator will be

closer to the true expected loss than other estimates
Robust = the estimator behaves well even if model assumptions are not
exactly met; stable results even given outliers
4
Basic Credibility Concepts Venters
Credibility for Dummies
Credibility theory that focuses on the goal of minimum variance is also
known as least squares or greatest accuracy credibility known as least squares or greatest accuracy credibility.
The goal is simple to state: We want to make use of all the
available and relevant information giving the proper weight available and relevant information, giving the proper weight
to each piece of information.
Credibility theory is all about weighted averages.
-Gary Venter
5
Basic Credibility Concepts Venters
Credibility for Dummies
A credibility-weighted (cw) average of two estimators is given as a
linear weighted average: linear weighted average:
cw 1 2

The two estimators are unbiased and independent:

1 2

The variance of the credibility-weighted average is written as:
Cov
1 2
=0
y g g
2
1
2
2 cw 1 2

6
Basic Credibility Concepts Venters
Credibility for Dummies
We can find the best credibility weight as the w that minimizes the
variance of the credibility weighted average variance of the credibility-weighted average.
cw
The result is that the best weight is inversely proportional to the
1

variance of the estimator.
1
1
2
1 2
1
-1
1
-1
2
-1

7
Basic Credibility Concepts
8
Multivariate Case
A credibility-weighted (cw) average of multiple estimators:
cw i i
n

i
n

If all estimators are assumed to be unbiased and independent:

i=1 i=1
p
n
cw i
2
i
i=1
9
Multivariate Case
Assuming independence among the various estimators, the best
weights are again inversely proportional to the individual weights are again inversely proportional to the individual
variances.
i
-1
i
]
-1
n
] =1
Substituting these weights back into the variance equation produces
the following:
cw i

1 2 n
10
Multivariate Case
Where there is correlation between the estimators, we define a
covariance matrix containing the covariance between every pair of covariance matrix containing the covariance between every pair of
estimators.
For the three variable case we have: For the three variable case, we have:
1 1 2 1 S
2 1 2 2 S 2 1 2 2 S
S 1 S 2 S
11
Multivariate Case
The weights to be applied to the estimators are represented as a
vector of numbers
I
vector of numbers.
The best value for the weights, constrained so that they sum to
unity is found by matrix operations
1 2 n
I

unity, is found by matrix operations.

-1
n
I
1

This is calculated by taking the inverse of the covariance matrix and
n
I
-1
n
then dividing each column total by the overall total.
12
Multivariate Case
Interesting Tangent:
The math is equivalent to minimum variance portfolio optimization.
Portfolio Asset Allocation Credibility Weights
CD's 3
T-Bills
Stocks
2
Bonds
1
13
The XOL Reinsurance Problem
The experience rate is an estimator of the future loss.
cxpcr

With variance based on:
Number of years and losses in the historical period
Attachment Point and Limit of layer being priced
Changing operations of the client company
The exposure rate is an estimator of the future loss.
With variance based on:
cxpos

Volume of loss experience in the industry
Relevance of industry experience to a specific client
14
Traditional Credibility Weighted Average
Example of Standard Credibility Procedure
2 000 000
Layer 2
1M 1M
Layer 2
1M 1M
Layer 2
1M 1M
2,000,000
w + w) =
1,000,000
1Mxs1M 1Mxs1M 1Mxs1M
)
500,000
Layer 1
500xs500
Layer 1
500xs500
Layer 1
500xs500
Experience Rating Exposure Rating Credibility-Weighted
15
The XOL Reinsurance Problem
Our goal is to produce an unbiased, minimum variance estimator of
the expected loss in the prospective period the expected loss in the prospective period.
The traditional credibility weighting can bring us part of the way, but it
does not make use of all the available information Namely the does not make use of all the available information. Namely, the
experience in lower layers is ignored.
An additional estimate can be produced using exposure-rating An additional estimate can be produced using exposure-rating
relativities applied to a lower layer (e.g., 500,000 xs 500,000).
rcl cxpcr _SuuxSuu
cxpos _1Hx1H
cxpos _SuuxSuu

16
Estimating Higher Layer based on Exposure-Rating
Relativities Applied to Lower Layer pp y
Using Exposure-Rating Relativities
2 000 000 2,000,000
Layer 2
1M 1M
Layer 2
1M 1M
Layer 2
1M 1M
=
1,000,000
1Mxs1M 1Mxs1M 1Mxs1M
500,000
Layer 1
500xs500
Layer 1
500xs500
Layer 1
500xs500

Experience Rating Exposure Ratings Lower Layer times Relativity

17
Calculating Variances - For Numerical Example
For a numerical example in the paper, we estimate variances for the
three methods: three methods:
Experience Rate based on loss volume in historical period
(ignores uncertainty for changing exposures etc) (ignores uncertainty for changing exposures, etc)
Exposure Rate based on uncertainty in Pareto distribution used
for size-of-loss and on uncertainty in overall frequency for size-of-loss and on uncertainty in overall frequency
Relativity Method based on Pareto distribution in size-of-loss
curve and on variance of experience rate for lower layer curve and on variance of experience rate for lower layer
Note: The covariances between the methods are set by the structure
of the model and do not have to be separately estimated.
18
Calculating Variances - Covariance Matrix
Exposure Experience Relativity
Covariance 1 573E+11 0 3 790E+10 Covariance 1.573E+11 0 3.790E+10
Matrix: 0 1.716E+11 7.322E+10
3.790E+10 7.322E+10 8.788E+10
Inverse: 7.580E-12 2.165E-12 -5.073E-12
2 165E-12 9 663E-12 -8 986E-12 2.165E 12 9.663E 12 8.986E 12
-5.073E-12 -8.986E-12 2.105E-11
Row Total: 4.672E-12 2.843E-12 6.996E-12
Weights: 32.2% 19.6% 48.2%
Total Variance: 6.891E+10
19
Recursive Credibility Form
The result is a three-factor credibility formula.
cw 1 cxpos _1Hx1H
2 cxpcr _1Hx1H

We can rearrange this expression into a recursive form:

S rcl

g p
p
cw
SuuxSuu
= _
w
1
w
1
+ w
S
] p
cxpo s
SuuxSuu
+ _
w
S
w
1
+ w
S
] p
cxpc r
SuuxSuu
_
w
1
+ w
S
] _
w
1
+ w
S
]

( ) _
p
cxpo s
1Hx 1H
_ p
cw
1Hx 1H
= (w
1
+ w
S
) p
cw
SuuxSuu
_
cxpo s
1Hx 1H
p
cxpo s
SuuxSuu
_ + w
2
p
cxpcr
1Hx 1H

20
Recursive Credibility Form
Numerical Example
Alternative Recursive Form
Experience Rating Exposure Rating Credibility-Weighted
Loss Cost Cred% Loss Cost Relativity Cred% Loss Cost Cred%
500 xs 500 5 000 000 60 0% 4 000 000 1 000 40 0% 4 600 000 100 0% 500 xs 500 5,000,000 60.0% 4,000,000 1.000 40.0% 4,600,000 100.0%
1M xs 1M 4,000,000 3,000,000 0.750
E i R ti C l t f C dibilit C dibilit W i ht d Experience Rating Complement of Credibility Credibility-Weighted
Loss Cost Cred% Loss Cost Relativity Cred% Loss Cost Cred%
500 xs 500 5,000,000 4,600,000 1.000
1M xs 1M 4,000,000 19.6% 3,450,000 0.750 80.4% 3,557,800 100.0%
21
Numbers for illustration only
Recursive Credibility Form
The recursive form of this credibility formula is already commonly
used; known as decay method or layer comparison used; known as decay method or layer comparison.
And the end of all our exploring
Will be to arrive where we started
And know the place for the first time p
- T.S. Eliot
22
THANK YOU VERY MUCH FOR YOUR
ATTENTION ATTENTION.
Dave Clark
Copyright 2011 Munich Reinsurance America, Inc. All rights reserved. "Munich Re" and the Munich Re logo are
internationally protected registered trademarks. The material in this presentation is provided for your information only, and y p g p p y y,
is not permitted to be further distributed without the express written permission of Munich Reinsurance America, Inc. or
Munich Re. This material is not intended to be legal, underwriting, financial, or any other type of professional advice.
Examples given are for illustrative purposes only. Each reader should consult an attorney and other appropriate advisors
to determine the applicability of any particular contract language to the reader's specific circumstances.