ASTAM

Updated 04/17/23

Severity & Aggregate

SEVERITY AND Models
AGGREGATE MODELS
Considerations when estimating GEV
If e- (d) has a negative slope, then ξ < 0,
• X is in Weibull EV MDA, and
parameters:
Mixtures • excess loss follows beta distribution.
• Fewer but larger blocks
Special Mixtures
o More volatile estimates
If ( N ∣ λ ) ∼ Poisson (λ) Misc. Topics on Aggregate Models
o Block maxima are more likely in the
where Λ ∼ Gamma (α, θ), Impact of Deductibles on Claim Frequency
tail
then N ∼ Neg. Binomial (α, θ). For v = Pr(X > d),
• More but smaller blocks
If ( X ∣ θ ) ∼ Normal (θ, v) o More stable estimates 𝐍𝐍 𝐍𝐍′
where Θ ∼ Normal (µ, a), o Block maxima are more likely in the Poisson λ vλ
then X ∼ Normal (µ, v + a). tail
Binomial m, q m, vq
If ( X ∣ q ) ∼ Bernoulli (q) Points Over Threshold (POT) Approach Neg. Binomial r, β r, vβ
where Q ∼ Beta (a, b), Generalized Pareto Distribution (GPD):
!
then X ∼ Bernoulli D E. For X ∼ GPD (ξ, β),
!"#
1
• if ξ < 1, then E[X] = . Sum of Compound Poisson Models
/&0
Splices
• (X − d ∣ X > d) ∼ GPD (ξ, β + ξd). A collective risk model where the
Note: The coefficients may not sum to 1.
frequency follows a Poisson distribution.
To determine the coefficients: Empirical Mean Excess Loss:
• Integrate/sum probability function and ∑% 34 x2 Discretization
e_(d) = # −d
equate to 1 n4 Let Y be the discretized version of X, and h
• Equate PDF at splicing point if PDF is be the span.
Risk Measures:
continuous
Pr(X > x) = S- (d) ⋅ S5 (x − d) Method of Rounding:
Frailty Models Pr(Y = y) = Pr(y − 0.5h < X ≤ y + 0.5h)
0
h( x ∣ Λ ) = Λ ⋅ a(x) ⎧d + β YfS- (d)h − 1i , ξ ≠ 0 for y = 0, h, 2h, …
⎪ ξ 1−p
S(x) = M$ [−A(x)], where A(x) = π6 = Method of Local Moment Matching:
⎨ S- (d)
%
∫&' a(t) dt ⎪ d + β ln f h, ξ=0 E[X ∧ h]
⎩ 1−p Pr(Y = 0) = 1 −
h
π6 + β − ξd and, for y = h, 2h, …,
Extreme Value Theory For ξ < 1, EkX ∣ X > π6 l =
1−ξ Pr(Y = y)
Tail Weight
2E[X ∧ y] − E[X ∧ y ∗& ] − E[X ∧ y ∗" ]
1. Fewer positive raw moments ⟹ The Hill Estimator: =
h
heavier tail ∑.297 ln x(2)
&/
where y ∗& = y − h and y ∗" = y + h.
2.
) (%)
If lim ! (%) =
, (%)
∞ or lim ! (%) = ∞, then m8
α7 =n − ln x(7) p
%→' )" %→' ,"
n−j+1
numerator has a heavier tail.
;%
&:
3. h(x) decreases with x ⟹ heavy tail n−j x $

Sq8 (x) = n p
4. e(d) increases with d ⟹ heavy tail n x(7)

Block Maxima Approach Relationship Between GEV and GPD

y
Generalized Extreme Value (GEV) G0,1 (y) = 1 + ln H0 t u
β
Distribution:
If e- (d) has a positive slope, then 0 < ξ < 1,
• Gumbel: ξ = 0
• X is in Fréchet MDA, and
• Fréchet: ξ > 0
• excess loss follows Pareto distribution.
• Weibull EV: ξ < 0
If e- (d) has a slope of 0, then ξ = 0,
X’s distribution is in the MDA of the GEV • X is in Gumbel MDA, and
distribution if and only if • excess loss follows exponential
/
& distribution.
lim n ⋅ S- (c. x + d. ) = Y(1 + ξx) 0, ξ≠0
%→'
e&% , ξ=0

© 2023 Coaching Actuaries. All Rights Reserved www.coachingactuaries.com ASTAM Formula Sheet 1
 CONSTRUCTION
ConstructionAND&SELECTION
SelectionOF of Hypothesis Tests Chi-Square Goodness-of-Fit Test Properties
PARAMETRIC MODELS
Parametric Models HB : null hypothesis • Individual and grouped data
H/ : alternative hypothesis • Continuous and discrete fit
Variance of MLE Reject HB when • No adjustments to critical value for
Fisher’s Information • test statistic > critical value or censored data
One Parameter: • p-value ≤ significance level • If parameters are estimated, critical
I(θ) = −E- [l′′(θ)] value is automatically adjusted via
𝐇𝐇𝟎𝟎 is true 𝐇𝐇𝟎𝟎 is false degrees of freedom
Varkθql = [I(θ)]&/
• No change for critical value if sample
Two Parameters: Type I Correct size is large
Reject 𝐇𝐇𝟎𝟎
l>>
: (α, θ) l>:,? (α, θ) Error Decision • Data needs to be grouped according to E7
I(α, θ) = −E- f > h
l:,? (α, θ) l>>? (α, θ) • More weights on intervals with poor fit
Fail to Correct Type II
m] Covkα
Var[α m, θql reject 𝐇𝐇𝟎𝟎 Decision Error
[I(α, θ)]&/ = n p Hypothesis Tests: Likelihood Ratio
m, θl Varkθql
Covkα q
Test statistic: T = 2[l(θ/ ) − l(θB )]
Hypothesis Tests: Kolmogorov-Smirnov Degrees of freedom
Delta Approximation Kolmogorov-Smirnov Test = # of free parameters in H/
One-Variable:
Test statistic: D = maxÄéDÄx7 Åé, éDÄx7& ÅéÅ − # of free parameters in HB
@ !DD 7
d
VarkgÄθqÅl ≈ É g(θ)Ñ Varkθql If data is left-truncated at d, then
dθ Score-Based Approaches
F(x) − F(d)
Two-Variable: F ∗ (x) = , for x ≥ d • Smallest K-S test statistic
1 − F(d)
m, θqÅl ≈ (g >: )@ Var[α
VarkgÄα m] If data is right-censored at u, then F. (u) is • Smallest chi-square test statistic
m, θql
+ 2g >: g >? Covkα indeterminate. • Largest chi-square p-value
+ (g ? ) Varkθql
> @ • Largest maximized likelihood
Kolmogorov-Smirnov Test Properties • Largest SBC/BIC
Normal Confidence Interval • Individual data only • Largest AIC
• Continuous fit only
à kθql
θq ± z(/"6)/@ áVar
• Lower critical value for censored data
r
• If parameters are estimated, critical value SBC/BIC l − ln n
2
Graphical Analyses should be adjusted
D(x) Plot AIC l−r
• Lower critical value if sample size is large
Graph the difference between empirical • No discretion where
CDF and fitted CDF • Uniform weight on all parts l: log-likelihood
of distribution r: # of estimated parameters
n: sample size
Hypothesis Tests: Chi-Square
Chi-Square Goodness-of-Fit Test
2 @
ÄE7 − O7 Å
Test statistic: χ@ = í
E7
79/
2
O@7
=í −n
E7
79/

Peak: DÄx7 Å = F. Äx7 Å − F Äx7 Å ∗ where

Valley: DÄx7& Å = F. Äx7& Å ∗
− F Äx7 Å • k: # of groups
# of observations ≤ x • E7 : expected # of observations in group j
where F. Äx7 Å =
n • O7 : actual # of observations in group j
& # of observations < x Degrees of freedom = k − 1 − r where
and F. Äx7 Å =
n • r: # of estimated parameters
p-p Plot
Coordinates: DF. Äx7 Å, F ∗ Äx7 ÅE
# of observations ≤ x
where F. Äx7 Å =
n+1

© 2023 Coaching Actuaries. All Rights Reserved www.coachingactuaries.com ASTAM Formula Sheet 2
 CREDIBILITY Credibility Bühlmann Credibility Exponential/Gamma
Expected Hypothetical Mean (EHM): /
• Model: ( X ∣ θ ) ∼ Exponential D E
?
Bayesian Credibility µ = EkE[X ∣ θ]l
• Prior: θ ∼ Gamma (α, β)
Model Distribution Expected Process Variance (EPV):
Distribution of model conditioned on a v = EkVar[X ∣ θ]l Posterior
parameter Variance of Hypothetical Mean (VHM): ( θ ∣ data ) ∼ Gamma (α∗ , β∗ )
Model density function: f( x ∣ θ ) a = VarkE[X ∣ θ]l
∗
• α =α+n
v • β∗ = (β&/ + ∑.G9/ xG )&/
Prior Distribution Bühlmann k: k =
a Predictive
Initial distribution of the parameter n
Prior density function: π(θ) Bühlmann Credibility Factor: Z = ( X ∣ data ) ∼ Pareto Äα = α∗ , θ = β∗ &/ Å
n+k
Bühlmann Credibility Premium:
Posterior Distribution Normal/Normal
PF = Zxû + (1 − Z)µ
Revised distribution of the parameter • Model: ( X ∣ θ ) ∼ Normal (θ, v)
= µ + Z(xû − µ)
Posterior density function: • Prior: θ ∼ Normal (µ, a)
Note: 𝑍𝑍 and 𝑥𝑥̅ have the same 𝑛𝑛.
f( data ∣ θ ) ⋅ π(θ)
π(θ ∣ data) = ' Posterior
∫&' f( data ∣ θ ) ⋅ π(θ) dθ
Bühlmann As Least Squares ( θ ∣ data ) ∼ Normal (µ∗ , a∗ )
Note: Use numerator and domain to check if Estimate of Bayesian • µ∗ = Zxû + (1 − Z)µ
(𝜃𝜃 ∣ 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑) follows a distribution in the exam
õ%Å@ § where
Minimize í £p% ÄY% − Y • a∗ = (1 − Z)a
tables to skip integration.
!DD % Predictive
Predictive Distribution Y% : Bayesian estimate given X/ = x µ = µ∗ ,
( X ∣ data ) ∼ Normal t @ u
Revised unconditional distribution õ
Y% : Bühlmann estimate given X/ = x σ = v + a∗
(w.r.t. parameter) of the model
Properties of a Bayesian/Bühlmann graph
Predictive density function: f(x ∣ data)
• Bühlmann estimates are on a straight line Exact Credibility
Predictive Mean = Bayesian Premium
• Bayesian estimates are within the range Bayesian estimate = Bühlmann estimate
Loss Function of hypothetical means • Poisson/Gamma
• There are Bayesian estimates above and • Binomial/Beta
Function Posterior below the Bühlmann line • Exponential/Inv. Gamma
to õ, 𝛉𝛉Å
𝐥𝐥Ä𝛉𝛉 Estimation, • Bühlmann estimates are between the • Normal/Normal
Minimize õ
𝛉𝛉 sample mean and theoretical mean
Empirical Bayes
Squared- @ Posterior Conjugate Priors
Äθq − θÅ Non-Parametric Methods
error loss mean Poisson/Gamma Uniform Exposures
• Model: ( X ∣ λ ) ∼ Poisson (λ) ∑HG9/ ∑.79/ xG7
Absolute Posterior µ_ = = xû
éθq − θé • Prior: λ ∼ Gamma (α, θ) r⋅n
loss median
@
Posterior ∑HG9/ ∑.79/ÄxG7 − xûG Å
Zero-one 1 for θq ≠ θ Posterior v_ =
( λ ∣ data ) ∼ Gamma (α∗ , θ∗ ) r(n − 1)
loss 0 for θq = θ mode
• α = α + ∑.G9/ xG
∗

/ &/ ∑HG9/(xûG − xû)@ v_

• θ∗ = D + nE a_ = −
? r−1 n
Predictive
Non-uniform Exposures
( X ∣ data ) ∼ Neg. Bin. (r = α∗ , β = θ∗ ) .
∑HG9/ ∑79/
&
mG7 xG7
µ_ = = xû
Binomial/Beta m
• Model: ( X ∣ q ) ∼ Binomial (m, q) . @
∑HG9/ ∑79/
&
mG7 ÄxG7 − xûG Å
• Prior: q ∼ Beta (a, b, 1) v_ =
∑HG9/(nG − 1)
Posterior
∑HG9/ mG (xûG − xû)@ − v_(r − 1)
( q ∣ data ) ∼ Beta (a∗ , b∗ , 1) a_ =
m − m&/ ∑HG9/ m@G
• a = a + ∑.G9/ xG
∗

• b∗ = b + [n(m) − ∑.G9/ xG ] Credibility premium for a risk in Class i:

Predictive õF = ZqG xûG + Ä1 − ZqG ŵ_
P
− Note: 𝑍𝑍¶I and 𝑥𝑥̅I have the same 𝑛𝑛.

© 2023 Coaching Actuaries. All Rights Reserved www.coachingactuaries.com ASTAM Formula Sheet 3
 Balancing the Estimators Discounting Testing for Calendar Year Effects
∑HG9/ ZG xûG Discount projected claims from midyear of 1. For each DY j, categorize fG,7 into S, L, or *.
Estimate EHM as: µ_ = H
∑G9/ ZG payment to the end of CY I. 2. For each CY k, determine S2 and L2 .
N .# &/
3. Calculate n2 = S2 + L2 and m2 = ∞ ±.
Empirical Bayes õG = í õ
R XG,7 (1 + 𝑟𝑟)&(I"M&O&B.Q) @
4. Z2 = min(S2 , L2 )
Semi-Parametric Methods 79J&G"/
n2 n − 1 n2
To estimate v: E[Z2 ] = − t 2 u .#
Tail 2 m2 2
1 n2 (n2 − 1)
Model 𝐯𝐯_ βq7 = ; βq = 1 Var[Z2 ] =
qλ7 N 4
Poisson xû n − 1 n2 (n2 − 1)
γ_B = 0; γ_7 = βq7 − βq7&/ −t 2 u
m2 4
Neg. Binomial with
xû(1 + β) Inflation + E[Z2 ] − E[Z2 ]@
fixed β
1. Eliminate the impact of inflation. 5. Z = ∑J&/
29/ Z2
Binomial with fixed q xû(1 − q) QJ J&/
!
XG,7 = XG,7 ⋅ , for i + j ≤ I E[Z] = í E[Z2 ]
Gamma with fixed θ xûθ QG"7
29/
2. Develop the claims. J&/
To estimate µ and a, use the non-parametric 3. Add back the impact of the expected Var[Z] = í Var[Z2 ]
method formulas shown above. inflation. 29/
QG"7 6. Perform a two-tailed test based on Z ∼
õG,7 = X
X õG,7
!
⋅ , for i + j > I
QJ Normal(E[Z], Var[Z])
RESERVING FOR SHORT-TERM 4. Estimate the outstanding claims.
Reserving for Short-Term
INSURANCE COVERAGE Testing the CL Assumptions The Bornhuetter-Ferguson Method
Insurance Coverage
Testing for Correlated Development Factors C≥G,N = CqG,N ⋅ βqJ&G + Ä1 − βqJ&G ŵG
Outstanding claims reserves (OCR) Pearson Correlation: where µG = EPG ⋅ ELR G
The amount of funds the insurer has set For each j, calculate test statistic
aside for losses that have been incurred but C≥G,7"/ = CqG,7 + Äβq7"/ − βq7 ŵG
n7 − 2
have not been settled. T7 = r7 ≠
1 − r7@
The Bühlmann-Straub Credibility Model
Possible reasons of delay: where r7 is the sample correlation CqG,N
W)
= ZG CqG,N + (1 − ZG )µ_
• delays in reporting coefficient between consecutive vectors of where
• claims processing delays fG,7 and n7 is the sample size.
∑JG9/ ZG CqG,N
• legal proceedings µ_ =
Then, perform a two-tailed test based on ∑JG9/ ZG
Main components of OCR: Student’s t-distribution with v7 = n7 − 2
J&/
1
• Reported but not settled (RBNS) claims v_ = í sG@
degrees of freedom. I
G9B
• Incurred but not reported (IBNR) claims
Spearman’s Rank Correlation: ∑JG9B mG ÄCqG,N − CûÅ − Iv
õ
a_ = / J
Repeat Pearson correlation, except instead m−X ∑G9B m@G
The Chain Ladder Method
1. Estimate the development factors of fG,7 , use their ranks in the vectors. βqJ&G
ZG =
qβJ&G + ;R
∑J&/&7
G9B CG,7"/ !Y
¶f7 = Combining 𝑇𝑇M ’s into one single test: J&/ @
∑J&/&7
G9B CG,7 R$ &@ 1 XG,7
∑R$ST T7 ⋅ R$ sG@ = í γ_7 n − CqG,N p
K&/
T= I−i γ_7
R &@ 79B
λq7 = ® ¶f2 ∑R$ST $R
$
mG = βqJ&G
L9M E[T] = 0 J
2. Estimate the ultimate claims ∑R$ST
R$ &@
R$ m = í mG
CqG,N = CG,J&G ⋅ λqJ&G Var[T] = G9B
R$ &@ "
∑R$ST U V
3. Estimate the outstanding claims R$ ∑JG9B CG,J&G
Cû = J
õ G = CqG,N − CG,J&G
R Perform a two-tailed test based on T ∼ ∑G9B mG
õ G = CG,J&G ÄλqJ&G − 1Å
R Normal(0, Var[T]).
CqG,N
W)@
= βqJ&G CqG,N + Ä1 − βqJ&G ÅCqG,N
W)
4. Calculate the outstanding claims reserve
J CqG,N
W)@
= ZG∗ CqG,N + (1 − ZG∗ )µ_
õ = íR
R õG where ZG∗ = 1 − Ä1 − βqJ&G Å(1 − ZG )
G9B

© 2023 Coaching Actuaries. All Rights Reserved www.coachingactuaries.com ASTAM Formula Sheet 4
 The Poisson Model PRICING FOR SHORT-TERM
Pricing for Short-Term Other Topics
CqG,N
Z
= CG,J&G + µ_G Ä1 − βqJ&G Å INSURANCE COVERAGE
Insurance Coverage Individual Risk Rating Plans
• Prospective
If µ_G is estimated the same way as the BF Ratemaking o Experience rating plans let the insurer
method, Poisson method = BF method. Loss Ratio Method modify an insured’s premium based on
LR + F their actual loss experience.
If the parameters are estimated using the Indicated Avg. Rate Change = −1
1 − V − Q^ actual experience
MLE, Poisson method = CL method. M = Zt u + (1 − Z)
expected experience
Indicated RelativityG
Frequency-Severity Models o Schedule rating plans let the insurer
LRG
= Current RelativityG ⋅ modify an insured’s premium based on
Simple way: LR W![_
the risk characteristics that have not
1. Apply the CL method to the frequency.
Indicated Base Rate = Current Base Rate been reflected in the past loss
C≥G,N
[ [
= CG,J&G ⋅ λq[J&G
1 + Indicated Avg. Rate Change experience.
2. Apply the CL method to the severity. ⋅
Off-Balance Factor o Composite rating plans let the insurer
C≥G,N
\ \
= CG,J&G ⋅ λq\J&G
Off-Balance Factor rate a policy with multiple coverages
3. Multiply the estimate ultimate frequency using a single exposure base, instead
Indicated Avg. Relativity
and severity to estimate the ultimate = of a separate exposure base for
Current Avg. Relativity
claims. each coverage.
C≥G,N = C≥G,N
[
⋅ C≥G,N
\
o Large deductible policies allow the
Pure Premium Method insureds to self-insure up to the
Preferred way:
Lû + E
∫` deductible amount.
1. Estimate the numbers claims settled in Indicated Avg. Rate =
1 − V − Q^ • Retrospective rating plans determine a
each DY.
policy's premium using the actual loss
¥ [
XG,7 = γµ[7 ⋅ CqG,N
H
Avg. RateG
Avg. RelativityG = experience of the same policy period.
2. Estimate the average costs per claim Base RateG
settled in each DY. Limited Average Severity
LG
∑G9B XG,7\ J&7 Adj. LûG = ∑. min(xG , u)
¥
X7\ = Avg. RelativityG ⋅ ExposureG LAS(u) = G9/
I−j+1 n
3. Multiply the two together to estimate the Adj. LûG k∑%&ab xG l + u ⋅ nb
Indicated RelativityG = =
outstanding claims. Adj. LûW![_ n
where nb is the number of losses greater
¥G,7 = X
X ¥G,7
[ ¥\
⋅ X7 Indicated Base Rate than u.
Indicated Avg. Rate Note: 𝐿𝐿𝐿𝐿𝐿𝐿(𝑢𝑢) is the empirical version of
=
Mack’s Model Indicated Avg. Relativity 𝐸𝐸[𝑋𝑋 ∧ 𝑢𝑢]
∑J&/&7 ¶ @
G9B CG,7 ÄfG,7 − f7 Å Credibility-Weighted Relativities
m@7
σ = Increased Limit Factor
I−1−j
Z(Indicated Relativity) LAS(u) + RLb
]
σN&@
m ILF =
If I = J, then m @
σN&/ = min nσ@
mN&@ ,m @
σN&T , +(1 − Z)(Current Relativity)
@ p LAS(b) + RL#
σN&T
m
• b: base-level limit
N&/
m@7
σ • u: increased limit
1 1
õ G ∣ 𝒟𝒟J Å ≈
MSEPÄR CqG,N
@
í @ n + p Note: Risk load (RL) function will be provided
f¶
79J&G 7
CqG,7 S7
by the question if needed.
J
Indicated Relativity = ILF
õ ∣ 𝒟𝒟J Å ≈ í MSEPÄR
MSEPÄR õ G ∣ 𝒟𝒟J Å
G9/ Loss Elimination Ratio
J&/ N&/ J
σ@7
m LAS(d) − LAS(b)
+2 í CqG,N
@
í @ ∏ í CqG,N π LER4 =
f¶7 S7 xû − LAS(b)
G9/ 79J&G D9G"/
• b: base-level deductible
J&7&/ • d: increased deductible
where SqG,7 = ∑G9B CG,7 .

Indicated Relativity = 1 − LER

© 2023 Coaching Actuaries. All Rights Reserved www.coachingactuaries.com ASTAM Formula Sheet 5
 FAM-S

INSURANCE AND REINSURANCE

Insurance and Fire Related products:
COVERAGES Standard fire policy (SFP) covers direct • Comprehensive major medical: covers
Reinsurance Coverages
loss from fire and lightning, and at least one small frequent and large infrequent
Short-Term Insurance Coverages of the following: costs; small deductibles
What Makes a Risk Insurable? • Personal coverage • Catastrophic major medical: covers
1. Economically feasible. • Commercial coverage expenses higher than regular major
2. Economic value of insurance • Increased covered perils medical expenses; very high deductibles
is calculable. • Increased covered loss • Short-term medical: single limited term;
3. Loss is definite. simpler underwriting; lower insurer cost
4. Loss is random in nature. Marine
• High risk pool plan: insures people with
5. Exposures in any rate class Inland marine insurance covers:
no insurance due to pre-existing condition
are homogeneous. • Domestic shipments
• Consumer directed plan: high-deductible
6. Exposure units are spatially and • Instrumentalities of transportation and
plan + personal spending account
temporally independent. communication (bridges, tunnels, etc.)
o HSA can only be opened with a HDHP
• Personal property floater risks
o Tax-free contributions, earnings, and
Automobile
Liability withdrawals (for medical purposes)
• Third-party bodily injury and
property damage Low frequency, high severity claims with
Affordable Care Act (ACA):
• Medical payments (tort); Personal injury potentially high litigation costs:
• Individual mandate
protection (no-fault) • Claims-made basis: losses covered only if
(must have insurance)
• Unidentified, underinsured, uninsured reported during policy period; insurer
• Partially standardized insurance
motorists protection may offer tail coverage
• Plans must cover 10 EHBs
• Collision and other than collision • Occurrence: losses covered even if
• No cost sharing for preventive services
reported after policy period
Homeowners • Plans must meet AV metal level
Doctrine of proximate cause Major Medical • Plans must set out-of-pocket limit below
• A loss is only covered if a covered peril Provider networks: government limit
is the proximate cause of • Build own: common for carriers with • Plans sold on public exchanges must
a covered consequence. geographically concentrated insureds satisfy certain tests/standards
• Contract with networks for hire: common
Coinsurance Provision: Dental
for carriers with geographically
I Usually group policies because:
min tI, ⋅ Lu , I < cF diverse insureds
Claim Payment, P = Y cF • High frequency, low severity
min(I, L) , I ≥ cF Main types of health insurance plans: • Tax subsidy
• Health Maintenance Organization (HMO)
Workers Compensation Product design drivers:
Prior to the passage of workers comp laws, • Preferred Provider Organization (PPO)
• Induced utilization: over-utilize benefits
it was difficult for workers to get Insurance benefit calculations: because insurance coverage is present
compensation for injury/illness. Reasons: • Provider • Out-of-pocket limits • Accumulated untreated conditions: ability
• Doctrine of contributory negligence discounts • Maximum limits to postpone treatment until insurance
• Fellow-servant doctrine • Deductibles • Internal limits is present
• Assumption-of-risk doctrine • Coinsurance • Copays • Anti-selection: Individuals choose dental
coverage that benefits them the most
Workers can expect:
• ACA: pediatric dental coverage required
• Medical care benefits
• Disability income benefits
• Death benefits
• Rehabilitation services and benefits

© 2023 Coaching Actuaries. All Rights Reserved www.coachingactuaries.com FAM-S Formula Sheet 6
 Other Topics Reinsurance Law of Total Probability
Deductibles, d • Facultative: Used for ceding Pr(X = x) = E5 [Pr(X = x ∣ Y)]
Fixed dollar deductible: individual risks.
Law of Total Expectation
a.k.a. ordinary deductible or deductible • Treaty: Used for ceding all risks in a
0, X≤d E- [X] = E5 kE-[X ∣ Y]l
Y = max(0, X − d) = ¡ specific line or class of business.
X − d, X > d
• Quota share: Both parties share a Law of Total Variance
Fixed percentage deductible: percentage of the total risk. Var- [X] = E5 kVar- [ X ∣ Y ]l
d • Surplus share: Both parties share a + Var5 kE- [X ∣ Y]l
d, X ≤
D = max(d, δX) = √ δ percentage of the total risk above the
d retention limit. Independence
δX, X >
δ For independent X and Y,
• Excess of loss: The reinsurer is responsible
0, X≤d
⎧ d for the claim amounts exceeding the • Pr(X = x, Y = y)
⎪ X − d, d<X≤
Y= δ retention limit. = Pr(X = x) ⋅ Pr(Y = y)
⎨ d • E[g(X) ⋅ h(Y)] = E[g(X)] ⋅ E[h(Y)]
⎪
⎩(1 − δ)X, X>
δ
Empirical Distributions
Disappearing deductible: Severity,
SEVERITY, Frequency,
FREQUENCY, & and /
Each data point has a probability of .
d, X≤a Aggregate Models
AGGREGATE MODELS .
b−X Empirical CDF:
D = √d t u, a < X ≤ b
b−a Basics # of observations ≤ x
F. (x) =
0, X>b n
CDFs, Survival Functions, and
Empirical 100pcd percentile: π6 = x(⌈.6⌉)
Hazard Functions
0, X<d % ∑. xG
⎧ X − d, d≤X≤a Sample mean: xû = G9/
⎪ F(x) = Pr(X ≤ x) = ≈ f(t) dt n
Y= b−X &'
⎨ X − d tb − a u , a < X ≤ b ' Biased sample variance:
⎪ S(x) = Pr(X > x) = ≈ f(t) dt ∑.G9/(xG − xû)@
⎩ X, X>b % Var[X] =
n
f(x)
Franchise deductible: h(x) = ∑.G9/ xG@
0, X ≤ d
S(x) = − xû @
Y=¡ % n
X, X > d H(x) = ≈ h(t) dt = − ln S(x) Unbiased sample variance:
&'
Policy Limits, u ∑.G9/(xG − xû)@
⇒ S(x) = e &8(%) s@ =
X, X ≤ u n−1
Y = min(X, u) = ƒ n
u, X > u Moments = ⋅ Var[X]
'
n−1
Coinsurance, α E[g(X)] = ≈ g(x) ⋅ f(x) dx
Y = αX &' Severity Models
'
E[g(X)] = ≈ g′(x) ⋅ S(x) dx Incomplete Gamma Function
Deductible, Policy Limit, and Coinsurance B Γ(α; x) = 1 − Pr(N < α)
0, X≤d k cd moment: µ>2 = EkX 2 l ; µ/> = µ where N~Poisson(λ = x)
Y = Yα(X − d), d < X ≤ m
u, X>m k cd central moment: µ2 = Ek(X − µ)2 l
Uniform (a, b)
where Var[X] = σ@ = µ@
a+b (a − b)@
d: ordinary deductible (set to 0 if not Var[g(X)] = E[g(X)@ ] − E[g(X)]@ E[X] = Var[X] =
2 12
applicable) Covariance: Cov(X, Y) = E[XY] − E[X]E[Y]
u: policy limit (set to ∞ if not applicable) σ Special Distribution Shortcuts
Coefficient of variation: CV =
µ
α: coinsurance (set to 1 if not applicable)
µT µ] 𝐗𝐗 𝐗𝐗 − 𝐝𝐝 ∣ 𝐗𝐗 > 𝐝𝐝
u Skewness = T ; Kurtosis = ]
m: maximum covered loss, i. e. m = + d σ σ
α Pareto (α, θ) Pareto (α, θ + d)
Conditional Distributions
Pr(A ∩ B) Exponential (θ) Exponential (θ)
Pr(A ∣ B) =
Pr(B) Uniform (a, b) Uniform (0, b − d)
f- (x)
f-∣7f-f2 (x) = ,
Pr(j < X < k) Distributions Relationship
where j < x < k Is equivalent to

Gamma (1, θ) Exponential (θ)

Beta (1, 1, θ) Uniform (0, θ)

© 2023 Coaching Actuaries. All Rights Reserved www.coachingactuaries.com FAM-S Formula Sheet 7
 Sum of Independent Random Variables Deductibles, d Frequency Models
𝐗𝐗 𝐢𝐢 ∑𝐧𝐧𝐢𝐢9𝟏𝟏 𝐗𝐗 𝐢𝐢 Ordinary deductible: Distributions Relationship
0, X<d Is equivalent to
Gamma (αG , θ) Gamma (∑.G9/ αG , θ) Y l = (X − d)" = ¡
X − d, X ≥ d
Normal E[Y l ] = E[(X − d)" ] = E[X] − E[X ∧ d] Binomial (1, q) Bernoulli (q)
Normal ĵG , σ@G Å Ek(Y l )2 l = Ek(X − d)2" l
Ä∑.G9/ µG , ∑.G9/ σ@G Å Neg. Bin. (1, β) Geometric (β)
'

Roles of Parameters = ≈ (x − d)2 f(x) dx

4 Sum of Independent Random Variables
Shape parameter: affects the general '
= ≈ k(x − d)2&/ S(x) dx 𝐗𝐗 𝐢𝐢 ∑𝐧𝐧𝐢𝐢9𝟏𝟏 𝐗𝐗 𝐢𝐢
shape of the distribution 4
Scale parameter: parameter that is E[X ∧ d] Poisson (λG ) Poisson (∑.G9/ λG)
multiplied by a positive constant when Loss elimination ratio: LER =
E[X] Binomial (mG , q) Binomial (∑.G9/ mG , q)
a random variable is multiplied by that
same constant, while all other parameters Franchise deductible: Neg. Bin. (rG , β) Neg. Bin. (∑.G9/ rG , β)
0, X ≤ d
are unchanged Yl = ¡
X, X > d (a, b, 0) Class Property
• To scale any continuous distribution on E[Y l ] = E[(X − d)" ] + d ⋅ S(d) p. b
the exam table except lognormal, inverse = a + , for n = 1, 2, ⋯
p.&/ n
Gaussian, and log-t, scale the parameter θ Payment per Payment
• To scale a normal distribution, Y Z : Payment per payment Note: Included in exam table Appendix B.2.
X~Normal (µ, σ@ ) E[Y l]
E[Y Z ] = ; E[Y l ] = E[Y Z ] ⋅ S(d) Choosing from (a, b, 0) Class
⇓ S(d)
Two methods to fit data to an (a, b, 0)
cX~Normal (cµ, (cσ)@ ) class distribution:
• To scale a lognormal distribution, With ordinary deductible d,
• Method 1: Compare xû and s@
X~Lognormal (µ, σ@ ) E[Y Z ] = e(d) = E[ X − d ∣ X > d ]
2.#
E[(X − d)" ] • Method 2: Observe the slope of
⇓ =
.#'!

cX~Lognormal (µ + ln c, σ@ ) S(d)
Distribution Method 1 Method 2
Discrete Mixtures Special Shortcuts for e(d)
Poisson xû = s@ 0
f- (x) = ∑.G9/ wG ⋅ f-& (x) where ∑.G9/ wG = 1
𝐞𝐞(𝐝𝐝)
Binomial xû > s@ Negative
Bernoulli Shortcut 𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄 (𝛉𝛉) θ
Neg. Binomial xû < s@ Positive
𝑎𝑎, Probability = 𝑞𝑞
If 𝑋𝑋 = ¡ , b−d
𝑏𝑏, Probability = 1 − 𝑞𝑞 𝐔𝐔𝐔𝐔𝐔𝐔𝐔𝐔𝐔𝐔𝐔𝐔𝐔𝐔 (𝐚𝐚, 𝐛𝐛)
@ 2 Zero-Truncated Distributions
then Var[𝑋𝑋] = (𝑎𝑎 − 𝑏𝑏) 𝑞𝑞(1 − 𝑞𝑞)
1
θ+d p^. = p , for n = 1, 2, ⋯
𝐏𝐏𝐏𝐏𝐏𝐏𝐏𝐏𝐏𝐏𝐏𝐏 (𝛂𝛂, 𝛉𝛉) 1 − pB .
Severity Models w/ Coverage α−1
1
Modifications d Ek(N^ )2 l = EkN2l
S-P 𝐏𝐏𝐏𝐏𝐏𝐏𝐏𝐏𝐏𝐏𝐏𝐏 (𝛂𝛂, 𝛉𝛉) 1 − pB
Y l : Payment per loss α−1
Zero-Modified Distributions
Policy Limits, u The Ultimate Formula pm m ^
. = Ä1 − pB Åp. , for n = 1, 2, ⋯
X, X < u E[Y l ]
Yl = X ∧ u = ƒ 1 − pm
u, X ≥ u pm
B
m d . = p
Ek(Y l = Ek(X ∧ u)2 l
l )2 = α(1 + r) tE £X ∧ § − E ÉX ∧ Ñu 1 − pB .
1+r 1+r
b
Ek(Y l )2 l = ≈ x 2 f(x) dx + u2 ⋅ S(u) where Ek(Nm )2 l = Ä1 − pm ^ 2
B ÅEk(N ) l
B d: deductible (set to 0 if not applicable) 1 − pm
B
b
u: policy limit (set to ∞ if not applicable) Ek(Nm )2 l = EkN2 l
Ek(Y l )2
l = ≈ kx 2&/
S(x) dx 1 − pB
B α: coinsurance (set to 1 if not applicable)
E[X ∧ u] r: inflation rate (set to 0 if not applicable)
Increased Limit Factor, ILF = Aggregate Models
E[X ∧ b] u
m: maximum covered loss, i. e. m = + d Collective Risk Model
• 𝑏𝑏: original limit α
If S = ∑n
G9/ X G for independent N
• 𝑢𝑢: increased limit
and X, then:
• E[S] = E[N]E[X]
• Var[S] = E[N]Var[X] + Var[N]E[X]@

© 2023 Coaching Actuaries. All Rights Reserved www.coachingactuaries.com FAM-S Formula Sheet 8
 Normal Approximation PARAMETRIC ESTIMATION
Parametric Estimation INTRODUCTION TO CREDIBILITY
Introduction to Credibility
Approximate S as Normal (E[S], Var[S])
Maximum Likelihood Estimators Classical Credibility
Lognormal Approximation a.k.a. Limited Fluctuation Credibility
Steps to Calculating MLE
Approximate S as lognormal with
1. L(θ) = ∏ f(x) 4
3. l> (θ) = l(θ) Full Credibility
parameters found by equating E[S] and 4?
2. l(θ) = ln L(θ)
Var[S] to lognormal mean and variance 4. Set l> (θ) = 0 # of exposures needed for full credibility,
n_ :
Aggregate Payments Full credibility of aggregate claims:
(
Incomplete Data
S = ∑n l n
G9/ Y = ∑G9/ Y :
Z
z(/"6)⁄@ @
n_ = £ § ÄCV)@ Å
• E[S] = E[N]E[Y l ] = E[N> ]E[Y Z ] Left-truncated at d f(x)⁄S(d) k
l] l ]@
• Var[S] = E[N]Var[Y + Var[N]E[Y Right-censored at u S(u) # of claims needed for full credibility, n\ :
= E[N> ]Var[Y Z ] + Var[N> ]E[Y Z ]@ Full credibility of aggregate claims:
Grouped data on
Pr(a < X ≤ b) z(/"6)⁄@ @ σ@n
interval (a, b] n\ = £ § n + CV-@ p
Risk Measures k µn
Value-at-Risk (VaR) Special Cases • Full credibility of claim frequency: set
VaR 6 (X) = π6 = F-&/ (p)
CV-@ = 0
Distribution Shortcuts
Tail-Value-at-Risk (TVaR) • Full credibility of claim severity:
TVaR6 (X) = EkX ∣ X > VaR 6 (X)l Gamma, xû σ@n
θq = set =0
fixed α α µn
= VaR 6 (X) + ekVaR 6 (X)l
n\
µ_ = xû n\ = n_ ⋅ µn ⟺ n_ =
µn
𝐓𝐓𝐓𝐓𝐓𝐓𝐓𝐓𝐩𝐩 (𝐗𝐗)
Normal Partial Credibility
∑.G9/(xG − µ_)@
ϕÄz6 Å σ@ =
m Credibility premium: PF = Zxû + (1 − Z)M
Normal µ + σf h n
1−p = M + Z(xû − M)
∑.G9/ ln xG where
ΦÄσ − z6 Å µ_ = • M: manual premium
Lognormal E[X] ⋅ f h n
1−p Lognormal • Z: credibility factor/credibility
∑.G9/(ln xG − µ_)@
m@ =
σ n n′
n
Coherence Square Root Rule: Z = ≠ = ≠
n_ n\
ρ(X) is coherent if it satisfies the Poisson λq = xû
properties below: where
Binomial, xû • n: actual # of exposures
• Translation invariance: ρ(X + c) = q_ =
fixed m m • n′: actual # of claims
ρ(X) + c
• Positive homogeneity: ρ(cX) = c ⋅ ρ(X) Neg. Binomial, xû
βq =
• Subadditivity: ρ(X + Y) ≤ ρ(X) + ρ(Y) fixed r r
• Monotonicity: ρ(X) ≤ ρ(Y),
Uniform Distribution on (0, θ):
if Pr(X ≤ Y) = 1
VaR is not coherent because it • θq = max(x/ , x@ , … , x. )
fails subadditivity.
TVaR is coherent.

Tail Weight
Fewer positive raw moments ⟹
heavier tail

© 2023 Coaching Actuaries. All Rights Reserved www.coachingactuaries.com FAM-S Formula Sheet 9
 Pricing
PRICING and Reserving
AND RESERVING FOR Premium
for Short-Term
SHORT-TERM INSURANCE
Insurance Coverages
COVERAGES

Data Preparation for Ratemaking

Losses

Unearned premium for CY i:

PGr = PGs − PGt + PG&/
r

Extension of Exposures Method

Recalculates the premiums of historical
Incurred losses for CY i: LJG = LZG + R G − R G&/ policies under the current rate level
where R G is the reserves at the end of CY i
Parallelogram Method
Incurred losses for AY or PY i: LJG = LZG + R G
Calculates average factors to be applied to the
where R G is the reserves as of the
aggregate historical premiums to make them
valuation date
on-level
Expenses and Profit
Eq Ratemaking
Variable Expense Ratio: V =
P Loss Ratio Method
E`
Fixed Expense Ratio: F = LR + F
P Indicated Avg. Rate Change = −1
1 − V − Q^
Permissible Loss Ratio: PLR = 1 − V − Q^ ,
where Q^ is the target profit and Pure Premium Method
contingencies ratio a.k.a. Loss Cost Method
Lû + E
∫`
Indicated Avg. Rate =
1 − V − Q^

© 2023 Coaching Actuaries. All Rights Reserved www.coachingactuaries.com FAM-S Formula Sheet 10