Concept I Concept II Concept III

STAT 3058/6058 - Risk Modelling 2

Ratemaking (2)

Garry Khemka

S2, 2024

1 / 57
Concept I Concept II Concept III

Concept I

2 / 57
Concept I Concept II Concept III

Development and Trending

Recent data usually not available

Use of historical data
Adjustments to make data representative

3 / 57
Concept I Concept II Concept III

Loss Development Factors

Similar to what we learned in Run-off Triangles

Calculation of loss development factors fj ’s
Can be the BCL factors, arithmetic or geometric average, or the ratio
of losses

Aim is to calculate
Expected Ultimate Losses
Expected number of Claims

4 / 57
Concept I Concept II Concept III

Loss Development Factors

It is possible to have incurred loss development factors that are less

than 1
Allowance for distoring effects of a few large losses
Allowance for changes in benefit paid or policy coverage

5 / 57
Concept I Concept II Concept III

Trending - Timeline

Loss development factors are for historical data

Experience Period: The data available to the actuary, say calendar
year X − 2
Valuation Date: The date at which rates are set, say today in
calendar year X
Expected Effective Period: The period rates will be in effect for,
say calendar year X + 1
The actuary needs to decide on expected loss cost for the future
expected effective period
Hence adjustments are necessary
The adjustment process is called trending.

6 / 57
Concept I Concept II Concept III

Trending

Involves fiting curves

Recall regression analysis
We limit to linear or exponential trends

Key Considerations
Range of historical data used
Recall stability and responsiveness objectives
Must ignore seasonality and random noise

Level of aggregation, ideally homogeneous groups of policies

Data to trend
Pure Premium, or
Frequency and Severity

7 / 57
Concept I Concept II Concept III

Example

Consider the following results after loss development.

Accident Earned Ultimate Number of
Year Exposure Units Losses Incurred Claims
2016 9,083 2,519,000 492
2017 9,594 2,886,526 540
2018 9,399 3,290,618 586
2019 10,650 3,881,463 628
2020 10,610 3,807,102 648
Table 1: Ratemaking data.

8 / 57
Concept I Concept II Concept III

Example

From the data we can calculate

Accident Average Average Pure
Year Claim Frequency Claim Severity Premium
2016 0.0542 5,120 277.32
2017 0.0563 5,342 300.87
2018 0.0623 5,617 350.11
2019 0.0590 6,176 364.47
2020 0.0610 5,880 358.81
Table 2: Average Frequency, Average Severity and Pure Premium.

9 / 57
Concept I Concept II Concept III

Loss Trends

We want to estimate loss trends using

A linear model
Y = α1 + β1 X

An exponential model

ln(Y ) = α1 + β1 X

We can fit the models to frequency and severeity separately, or simply

to pure premium
We will illustrate using pure premium

10 / 57
Concept I Concept II Concept III

Example

Our regression data are

X = Year − AY 1 Y1 = PP Y2 = ln(PP)
0 277.32 5.625
1 300.87 5.707
2 350.11 5.858
3 364.47 5.898
4 358.81 5.883
Table 3: Data for curve fitting.

11 / 57
Concept I Concept II Concept III

Example

Linear model fit

Y = 284.9987 + 22.6589X

Exponential model fit

ln(Y ) = 5.6529 + 0.0707X

12 / 57
Concept I Concept II Concept III

Trending - Timeline revisited

Need to be careful to project loss data to an appropraiate mid-point

of expected effective period
This is called the loss trend period
Once a rate is set, it is in effect for a period of time (say one-year)
All policies sold in that period have the same loss trend in premium
calculations
If the policies are of duration one year, then the last policy sold will
be in effect for another year
Hence, the rate is in effect for two years and not one year!

13 / 57
Concept I Concept II Concept III

Trending - Timeline revisited

Consider:
The losses to be trended are from Accident Year 2011.
The company writes annual policies.
The proposed effective date is January 1, 2015.
The length of time the rates are expected to be in effect is one year.

14 / 57
Concept I Concept II Concept III

Trending - Timeline revisited

The average loss occurrence date of Calendar-Accident Year 2011 is

the midpoint, i.e., July 1, 2011
This is called the “trend from” date

The average loss occurrence date of the expected effective period is

December 31, 2015
This is called the “trend to” date
The loss trend period for Calendar-Accident Year 2011 is 4.5 years
This is what needs to be applied in the calculations
Not 4 years!

15 / 57
Concept I Concept II Concept III

Trending - Timeline revisited

The average loss occurrence date of Calendar-Accident Year 2011 is

the midpoint, i.e., July 1, 2011
This is called the “trend from” date

The average loss occurrence date of the expected effective period is

December 31, 2015
This is called the “trend to” date
The loss trend period for Calendar-Accident Year 2011 is 4.5 years
This is what needs to be applied in the calculations
Not 4 years!

Question
What would be the loss trend period if policies were semi-annual?

16 / 57
Concept I Concept II Concept III

Example continued
Estimate the expected loss cost for rates that take effect on
September 1 of 2021 on one-year policies. Assume that the rate
will be in effect for one year.
We have calendar year accident data for annual policies and rates will
be in effect for one year
The expected effective period in this case is September 1, 2022
Hence, the projected X is 6.1667
For linear model

Projected Pure Premium1 = 424.7284

For exponential model

Projected Pure Premium2 = 395.3867.

17 / 57
Concept I Concept II Concept III

Trending - Adjustments

Recent court adjudications (e.g., a change in the legal no-fault

threshold),
Interpretation of recent legislative changes (e.g., a change in speed
limits),
An expected change in the rate of economic inflation,
Level of economic activity (e.g., during periods of economic
downturns, workers compensation claims rise but auto collision claims
decline),
Recent changes in underwriting criteria or definitions, and
Changes to mandated benefit levels.

18 / 57
Concept I Concept II Concept III

Concept II

19 / 57
Concept I Concept II Concept III

Adjustment to Premiums - Current rate Level

Historical premiums need to be adjusted to current rate level
Also called on-level premiums
Allowances are to ensure that all historical premiums are at the same
level for projections
Illustration:
Original premiums were $200
At some point rates increased by 5% to $210
If true rates are $220, and the 5% increase is not taken into account,
then premiums should (incorrectly) increase by 10%
When 5% taken into account, rate increase is 4.8%

Two methods
Extension of exposures
Most accurate, but requires significant calculations and data

Parellelogram or geometric method

Less accurate
20 / 57
Concept I Concept II Concept III

Extension of Exposures

X
Earned Premium at Current Rates ($) = CRijk × eijk
ijk

where
CRijk is the current rate for cell ijk defined by rate classification
parameters i, j and k (for example, class, territory and rate group),
and
eijk is the corresponding number of earned exposure units in cell ijk in
the historical book of business.
[Note, the rate classification parameters come from issuing policies to
heterogeneous risks - Concepts IV]

21 / 57
Concept I Concept II Concept III

Example

Example
Assume that all policies have annual terms and premium is calculated according to the
following rating algorithm:

Premium = Exposure × Rate per Exposure × Class Factor + Policy Fee

The class factor has three values, or levels (X, Y, and Z), each with a distinct rate
differential.
The following three rate changes occurred during or after the historical experience
period:
July 1, 2010: The base rate was increased and this resulted in an overall average
rate level increase of 5%.
January 1, 2011: The base rate and policy fee were adjusted resulting in an overall
average rate level increase of 10%.
April 1, 2012: The policy fee and class Y and Z rate relativities were changed
resulting in an overall average rate level decrease of -1%.

22 / 57
Concept I Concept II Concept III

Example

Example
The details of each rate level are as follows:

You want to adjust the historical premium for Policy Year 2011 to the current rate level
for a policy effective on March 1, 2011 with 10 class Y exposures.
Actual premium charged: $7,370 (= 10 x $1,045 x 0.60 + $1,100)
On-level premium: $8,405 (= 10 x $1,045 x 0.70 + $1,090)
(You substitute the current base rate, class factor, and policy fee in the calculations
from actual to on-level)

23 / 57
Concept I Concept II Concept III

Concept III

24 / 57
Concept I Concept II Concept III

Credibility Theory
Two components to rating risks in experience rating plans
Classification rating plans: Rates based on characteristics or rating
factors
Insurers own experience: Based on historical loss experience

Insurer usues a credibility weight Z with 0 ≤ Z ≤ 1

R̂ = Z X̄ + (1 − Z )M,

R̂ = credibility weighted rate for risk,

X̄ = average loss for the risk over a specified time period,
M = the rate for the classification group, often called the manual rate.

Risks that have stable experience are likely to get weight Z close to 1
and vice versa
25 / 57
Concept I Concept II Concept III

Credibility Theory

Can be used for rate calculations within a classification rating plan

Some groups may have little or no data
Actual loss experience within group will have weight Z
Experience across all other classes will get the complement weight
1−Z

Can be used to update class rating plans

Weight 1 − Z to current rate plan

Can be applied to the calculation of expected frequencies and

severities
In addition to trending
Usually relates to manual updates due to external changes

26 / 57
Concept I Concept II Concept III

Limited Fluctuation Credibility

Limited fluctuation credibility (or classical or American

credibility) limits fluctuations in claim frequencies, severities or losses
Defines the minimum number of data points (usually claims) needed
for full credibility
We cover
Full and partial credibility
Claim frequency, severity, aggregate loss and pure premium

27 / 57
Concept I Concept II Concept III

Full Credibility for Claim Frequency

Let N be a random variable representing the number of claims for a group
of risks
Recall
pmf pN (n),
mgf mN (t),
first raw moment ν = E (N),
2nd central moment τ 2 = Var (N)
How big does ν need to be to get a good estimate?
We may find by estimating
Pr[(1 − k)ν ≤ N ≤ (1 + k)ν] ≥ p
The expected number of claims required for the probability on the left-hand
side to equal p is called the full credibility standard
Under full credibility Z = 1
Usually, ν is unknown, so estimate it by the observed number of
claims n
k and p are selected by the actuary
28 / 57
Concept I Concept II Concept III

Full Credibility for Claim Frequency

We need to estimate

Pr[(1 − k)ν ≤ N ≤ (1 + k)ν] ≥ p

We can standardize this to

 
−kν N −ν kν
Pr ≤ ≤ ≥p
τ τ τ

N−ν
We approximate τ by the standard normal distribution
If the claims are independent and N is large, then approximation is
reasonable

29 / 57
Concept I Concept II Concept III

Full Credibility for Claim Frequency

Define yp :

N −ν
Pr[−yp ≤ ≤ yp ] = Φ(yp ) − Φ(−yp ) = p
τ
−1
=⇒ yp = Φ ((p + 1)/2)

where Φ(·) is the cumulative distribution function of the standard normal

30 / 57
Concept I Concept II Concept III

N ∼ Poisson(λ)

Pr[N = n] = λn e−λ /n!

We know: λ = ν = τ 2
This implies kν/ν 1/2 ≥ yp
That is ν ≥ (yp /k)2
Let λkp to be the value of ν for which equality holds. Then full
credibility standard for the Poisson distribution is
 y 2
p
λkp = with yp = Φ−1 ((p + 1)/2)
k

If ν ≥ λkp then full credibility can be assigned to the data

Note, in practise, we use n ≥ λkp

31 / 57
Concept I Concept II Concept III

Example

Example
The full credibility standard is set so that the observed number of claims is
to be within 5% of the expected value with probability p = 0.95. If the
number of claims has a Poisson distribution find the number of claims
needed for full credibility.

32 / 57
Concept I Concept II Concept III

Example

Example
The full credibility standard is set so that the observed number of claims is
to be within 5% of the expected value with probability p = 0.95. If the
number of claims has a Poisson distribution find the number of claims
needed for full credibility.

Referring to a standard normal distribution table,

yp = Φ−1 ((p + 1)/2) = Φ−1 ((0.95 + 1)/2)=Φ−1 (0.975) = 1.960. Using
this value and k = 0.05 then λkp = (yp /k)2 = (1.960/0.05)2 = 1, 536.64.
After rounding up the full credibility standard is 1,537.

33 / 57
Concept I Concept II Concept III

N is not Poisson

We have kν/τ = yp
Squaring both sides and and rearranging

ν = (yp /k)2 (τ 2 /ν)

Full credibility standard for frequency

 y 2  τ 2   2
τ
p
nf = = λkp
k ν ν

34 / 57
Concept I Concept II Concept III

Using Exposures in Full Credibility for Claim

Frequency

Recall, exposure is a measure of risk

One car insured for a full year would be one car-year
Two cars each insured for exactly one-half year would also result in
one car-year

Doubling of exposure leads to doubling of (expected) claims

One-to-one relationship between exposure and claims

35 / 57
Concept I Concept II Concept III

Using Exposures in Full Credibility for Claim

Frequency
Need number of claims per exposure, which is frequency
Let m be the number of exposures then observed claim frequency is
N/m
We want to estimate E(N/m)

Pr[(1 − k)E(N/m) ≤ N/m ≤ (1 + k)E(N/m)] ≥ p

Noting, that E(N/m) = E(N)/m = ν/m

 
ν N ν
Pr (1 − k) ≤ ≤ (1 + k) ≥p
m m m

Now we can apply the full credibility standards developed before

36 / 57
Concept I Concept II Concept III

Example

Example
The full credibility standard should be selected so that the observed
number of claims will be within 5% of the expected value with probability
p = 0.95. The number of claims has a Poisson distribution. If one
exposure is expected to have about 0.20 claims per year, find the number
of exposures needed for full credibility.

37 / 57
Concept I Concept II Concept III

Example

Example
The full credibility standard is set so that the observed number of claims is
to be within 5% of the expected value with probability p = 0.95. If the
number of claims has a Poisson distribution find the number of claims
needed for full credibility.

With p = 0.95 and k = .05, λkp = (yp /k)2 = (1.960/0.05)2 = 1, 536.64

claims are required for full credibility. The claims frequency rate is 0.20
claims per exposure. To convert the full credibility standard to a standard
denominated in exposures the calculation is: (1,536.64 claims)/(0.20
claims/exposures) = 7,683.20 exposures. This can be rounded up to
7,684.

38 / 57
Concept I Concept II Concept III

Full Credibility for Aggregate Loss

Let S represent aggregate losses

S = X1 + X2 + · · · + XN .

where X1 , X2 , . . . , XN are iid

The mean and variance of S are

µS = E(S) = E(N)E(X ) = νµ1

and

σS2 = Var(S) = E(N)Var(X ) + [E(X )]2 Var(N) = νσ12 + µ21 τ 2 ,

39 / 57
Concept I Concept II Concept III

Full Credibility for Aggregate Loss

Aim to estimate expected losses µS = E(S) though observed losses S

Pr[(1 − k)µS ≤ S ≤ (1 + k)µS ] ≥ p

−kµS kµS
h i
=⇒ Pr ≤ (S − µS )/σS ≤ ≥p
σS σS

We assume that (S − µS )/σS is standard normal. Then we get

" 2 # " 2 #
2     
yp τ2 σ1 τ2 σ1

nS = + = λkp +
k ν µ1 ν µ1

[Note if claims are Poisson, this can be simplified to

[(τ 2 /ν) + (σ1 /µ1 )2 ] = [1 + (σ1 /µ1 )2 ] = [(µ21 + σ12 )/µ21 ] = µ2 /µ21

with full credibility standard nS = λkp µ2 /µ21 ]

40 / 57
Concept I Concept II Concept III

Example

Example
The number of claims has a Poisson distribution. Individual loss amounts
are independently and identically distributed with a Pareto distribution
F (x ) = 1 − [θ/(x + θ)]α . The number of claims and loss amounts are
independent. If observed aggregate losses should be within 5% of the
expected value with probability p = 0.95, how many losses are required for
full credibility?

41 / 57
Concept I Concept II Concept III

Example

Example
Because the number of claims is Poisson, (τ 2 /ν) = 1. The mean of the
Pareto is µ1 = θ/(α − 1) and the variance is σ12 = θ2 α/[(α − 1)2 (α − 2)]
so (σ1 /µ1 )2 = α/(α − 2). Combining the frequency and severity terms
gives [(τ 2 /ν) + (σ1 /µ1 )2 ] = 2(α − 1)/(α − 2). From a standard normal
distribution table yp = Φ−1 ((0.95 + 1)/2) = 1.960. The full credibility
standard is
nS = (1.96/0.05)2 [2(α − 1)/(α − 2)] = 3, 073.28(α − 1)/(α − 2).
Suppose α = 3 then nS = 6, 146.56 for a full credibility standard of 6,147.

Note that considerably more claims are needed for full credibility for
aggregate losses than frequency alone.

42 / 57
Concept I Concept II Concept III

Full Credibility for Pure Premium

The pure premium PP is equal to aggregate losses S divided by exposures
m: PP = S/m.
The full credibility standard for pure premium requires

Pr [(1 − k)µPP ≤ PP ≤ (1 + k)µPP ] ≥ p

As the number of exposures is assumed to be fixed, we have

µPP = E(S/m) = E(S)/m = µS /m. Hence

µS S µS
h      i
Pr (1 − k) ≤ ≤ (1 + k) ≥p
m m m
Since multiplying through by m leads to the expression for losses, this
implies

"   2 #
τ2 σ1
nPP = nS = λkp + .
ν µ1
43 / 57
Concept I Concept II Concept III

Full Credibility for Severity

Let X be a random variable representing the size of one claim

Claim severity is µ1 = E(X )
Suppose that X1 , X2 , . . . , Xn is a random sample of n iid claims that
will be used to estimate claim severity µ1
Sample average

1
X̄ = (X1 + X2 + · · · + Xn ) .
n
Note, n is not a random variable.
Aim: to find the minimum size of n for full credibility estimate

44 / 57
Concept I Concept II Concept III

Full Credibility for Severity

Requirement for number of claims
Pr[(1 − k)µ1 ≤ X̄ ≤ (1 + k)µ1 ] ≥ p
Standardisation gives
 
−k µ1 k µ1
Pr ≤ (X̄ − µ1 )/σX̄ ≤ ≥ p.
σX̄ σX̄
where, σX̄ is the standard deviation of the claim severity estimator, and
(X̄ − µ1 )/σX̄ is approximately normally distributed.
[Note, since X̄ is the average of individual claims X1 , X2 , . . . , Xn , its
standard deviation
√ is equal√to the standard deviation of an individual claim
divided by n: σX̄ = σ1 / n]
The full credibility standard for severity then is
 y 2  σ  2  2
σ1
p 1
nX = = λkp
k µ1 µ1
45 / 57
Concept I Concept II Concept III

Example

Example
Individual loss amounts are independently and identically distributed with
a Type II Pareto distribution F (x ) = 1 − [θ/(x + θ)]α . How many claims
are required for the average severity of observed claims to be within 5% of
the expected severity with probability p = 0.95?

46 / 57
Concept I Concept II Concept III

Example

Example
Individual loss amounts are independently and identically distributed with
a Type II Pareto distribution F (x ) = 1 − [θ/(x + θ)]α . How many claims
are required for the average severity of observed claims to be within 5% of
the expected severity with probability p = 0.95?

The mean of the Pareto is µ1 = θ/(α − 1) and the variance is

σ12 = θ2 α/[(α − 1)2 (α − 2)] so (σ1 /µ1 )2 = α/(α − 2). From a standard
normal distribution table yp = Φ−1 ((0.95 + 1)/2) = 1.960. The full
credibility standard is nX = (1.96/0.05)2 [α/(α − 2)] = 1, 536.64α/(α − 2).
Suppose α = 3 then nX = 4, 609.92 for a full credibility standard of 4,610.

47 / 57
Concept I Concept II Concept III

Relationship between full credibility standards

The full credibility standard for aggregate losses or pure premium is
" #
 y 2  τ 2   σ 2
p 1
nS = + ,
k ν µ1

If one assumes σ1 = 0 then we obtain the full credibility standard for

frequency results
If τ = 0 then the full credibility formula for severity follows
In practice, it is often assumed that the number of claims is Poisson
distributed so that τ 2 /ν = 1
Then, we have

 y 2  µ 
p 2
nS =
k µ21
48 / 57
Concept I Concept II Concept III

Partial Credibility

The full credibility stardards (let them be n0 ) denotes the minimum

number of claims required to achieve a defined level of accuarcy
Under full credibility, credibility weight is Z = 1 to the experience data
If n is the number of claims for a risk (or risk classification), the
credibility weights assigned to the data are
 p
n/n0 if n < n0
Z=
1 if n ≥ n0 ,

Note, the presence of the square root implies that variance (or
standard deviation) is used as the measure of fluctuation

49 / 57
Concept I Concept II Concept III

Example

Example
The number of claims has a Poisson distribution. Individual loss amounts
are independently and identically distributed with a Type II Pareto
distribution F (x ) = 1 − [θ/(x + θ)]α . Assume that α = 3. The number of
claims and loss amounts are independent. The full credibility standard is
that the observed pure premium should be within 5% of the expected
value with probability p = 0.95. What credibility Z is assigned to a pure
premium computed from 1,000 claims?

50 / 57
Concept I Concept II Concept III

Example
Example
Since the number of claims is Poisson,
2
E(X 2 ) τ2

σ1
= + .
[E (X )]2 ν µ1
The mean of the Pareto is µ1 = θ/(α − 1) and the second moment is
E(X 2 ) = 2θ2 /[(α − 1)(α − 2)] so E(X 2 )/[E (X )]2 = 2(α − 1)/(α − 2).
From a standard normal distribution table,
yp = Φ−1 ((0.95 + 1)/2) = 1.960. The full credibility standard is

nPP = (1.96/0.05)2 [2(α − 1)/(α − 2)] = 3, 073.28(α − 1)/(α − 2)

and if α = 3 then n0 = nPP = 6, 146.56 or 6,147 if rounded up. The

credibility assigned to 1,000 claims is Z = (1, 000/6, 147)1/2 = 0.40.

\end{block}
51 / 57
Concept I Concept II Concept III

Exam Style Question 1

Question
The full credibility standard for a company is set so that the total number
of claims is to be within 4% of the true number with probability P. This
full credibility standard is calculated to be 3500 claims.
The standard is now altered so that the aggregate claim amount is to be
within 10% of the true value with probability P.
The claim frequency has a Poisson distribution and the claim severity has
the following distribution:

f (x ) = 0.0003125(80 − x ), 0 ≤ x ≤ 80.

What is the expected number of claims necessary to obtain full credibility

under the new standard?

52 / 57
Concept I Concept II Concept III

Exam Style Question 1

Solution
Z 80

E (X ) = xf (x )dx
0
Z 80
2
= 0.0003125 (80x − x )dx
0
2 3 80
= 0.0003125(40x − x /3)|0 = 26.67.
Z 80
2 2
E (X ) = x f (x )dx
0
Z 80
2 3
= 0.0003125 (80x − x )dx
0
3 4 80
= 0.0003125((80/3)x − x /4)|0 = 1, 066.67.

53 / 57
Concept I Concept II Concept III

Exam Style Question 1

Solution
For N Poisson, we know from lectures:
yp 2
λkp = ( )
k
y p 2
Therefore, 3, 500 = ( 0.04 )
For aggregate claims, we know from lectures:

µ2 yp 2 E (X 2 )
ns = λkp =( )
µ21 0.04 E (X )2

1, 066.67
= 3, 500 = 3, 500(1.5) = 5, 250.
26.672
We need to adjust this for the new k = 10%:

0.042
ns = 5, 250 × = 840.
0.102

54 / 57
Concept I Concept II Concept III

Exam Style Question 2

Question
You are given the following:
The number of claims follows a Poisson distribution.
Claim severity follows a lognormal distribution with parameters µ and
σ.
The number of claims and the size of a claim are independent.
13, 000 claims are needed for full credibility for the aggregate claim
amount according to the limited fluctuation credibility approach.
The full credibility standard has been selected so that the actual
aggregate claim amount will be within 3% of expected aggregate
claim amount 90% of the time using a normal approximation.
Determine σ.

55 / 57
Concept I Concept II Concept III

Exam Style Question 2

Solution
For the lognormal distribution:

E (X ) = exp(µ + σ 2 /2)

E (X 2 ) = exp(2µ + 2σ 2 )
For aggregate claims when the number of claims is Poisson, we have from
lectures: ns = λkp µµ22 .
1
We are given in the question that k = 0.03 and p = 0.90. Therefore:
yp 2
λkp = ( 0.03 ) , where
yp = Φ ((p + 1)/2) = Φ−1 ((0.9 + 1)/2) = Φ−1 (0.95) = 1.645
−1

56 / 57
Concept I Concept II Concept III

Exam Style Question 2

Solution
Therefore:
mu2
13, 000 = λkp
mu12
1.645 2 exp(2µ + 2σ 2 )
13, 000 = ( )
0.03 exp(2µ + 2σ 2 /2)
1.645 2
13, 000 = ( ) exp(σ 2 )
0.03
Then solving for σ we get: σ = 1.21.

57 / 57