Exam STAM

updated 12/20/21

SEVERITY, FREQUENCY
SEVERITY, &
FREQUENCY & Conditional Distributions Special Distribution Shortcuts
AGGREGATE
AGGREGATE MODELS MODELS Pr(𝐴𝐴 ∩ 𝐵𝐵) Pr(𝐵𝐵 ∣ 𝐴𝐴) Pr(𝐴𝐴)
Pr(𝐴𝐴 ∣ 𝐵𝐵) = = 𝑿𝑿 𝑿𝑿 − 𝒅𝒅 ∣ 𝑿𝑿 > 𝒅𝒅
Pr(𝐵𝐵) Pr(𝐵𝐵)
Basic 𝑓𝑓0 (𝑥𝑥) Pareto (𝛼𝛼, 𝜃𝜃) Pareto (𝛼𝛼, 𝜃𝜃 + 𝑑𝑑)
𝑓𝑓0∣4505* (𝑥𝑥) = ,
CDFs, Survival Functions, and Pr(𝑗𝑗 < 𝑋𝑋 < 𝑘𝑘)
where 𝑗𝑗 < 𝑥𝑥 < 𝑘𝑘 Exponential (𝜃𝜃) Exponential (𝜃𝜃)
Hazard Functions
!
Uniform (𝑎𝑎, 𝑏𝑏) Uniform (0, 𝑏𝑏 − 𝑑𝑑)
𝐹𝐹(𝑥𝑥) = Pr(𝑋𝑋 ≤ 𝑥𝑥) = * 𝑓𝑓(𝑡𝑡) d𝑡𝑡 Law of Total Probability
"#
Pr(𝑋𝑋 = 𝑥𝑥) = E6 [Pr(𝑋𝑋 = 𝑥𝑥 ∣ 𝑌𝑌)]
# Uniform (𝑎𝑎, 𝑏𝑏)
𝑆𝑆(𝑥𝑥) = Pr(𝑋𝑋 > 𝑥𝑥) = * 𝑓𝑓(𝑡𝑡) d𝑡𝑡 𝑎𝑎 + 𝑏𝑏 (𝑎𝑎 − 𝑏𝑏)-
! Law of Total Expectation E[𝑋𝑋] = Var[𝑋𝑋] =
𝑓𝑓(𝑥𝑥) E0 [𝑋𝑋] = E6 fE0 [𝑋𝑋 ∣ 𝑌𝑌]g 2 12
ℎ(𝑥𝑥) =
𝑆𝑆(𝑥𝑥) Zero-Truncated Distributions
! Law of Total Variance
𝐻𝐻(𝑥𝑥) = * ℎ(𝑡𝑡) d𝑡𝑡 = − ln 𝑆𝑆(𝑥𝑥) 1
Var0 [𝑋𝑋] = E6 fVar0 [ 𝑋𝑋 ∣ 𝑌𝑌 ]g 𝑝𝑝2: = 𝑝𝑝 , for 𝑛𝑛 = 1, 2, ⋯
"# 1 − 𝑝𝑝' 2
⇒ 𝑆𝑆(𝑥𝑥) = 𝑒𝑒 "$(!) + Var6 fE0 [𝑋𝑋 ∣ 𝑌𝑌]g 1
E[(𝑁𝑁 : )* ] = E[𝑁𝑁 * ]
1 − 𝑝𝑝'
Moments Independence
#
For independent 𝑋𝑋 and 𝑌𝑌, Zero-Modified Distributions
E[𝑔𝑔(𝑋𝑋)] = * 𝑔𝑔(𝑥𝑥) ⋅ 𝑓𝑓(𝑥𝑥) d𝑥𝑥
"# • Pr(𝑋𝑋 = 𝑥𝑥, 𝑌𝑌 = 𝑦𝑦) 1 − 𝑝𝑝';
# 𝑝𝑝2; = 𝑝𝑝 , for 𝑛𝑛 = 1, 2, ⋯
= Pr(𝑋𝑋 = 𝑥𝑥) ⋅ Pr(𝑌𝑌 = 𝑦𝑦) 1 − 𝑝𝑝' 2
= * 𝑔𝑔′(𝑥𝑥) ⋅ 𝑆𝑆(𝑥𝑥) d𝑥𝑥
' • E[𝑔𝑔(𝑋𝑋) ⋅ ℎ(𝑌𝑌)] = E[𝑔𝑔(𝑋𝑋)] ⋅ E[ℎ(𝑌𝑌)] 1 − 𝑝𝑝';
E[(𝑁𝑁 ; )* ] = E[𝑁𝑁 * ]
𝑘𝑘 () moment: 𝜇𝜇*+ = E[𝑋𝑋 * ] ; 𝜇𝜇,+ = 𝜇𝜇 1 − 𝑝𝑝'
Empirical Distributions
𝑘𝑘 () central moment: 𝜇𝜇* = E[(𝑋𝑋 − 𝜇𝜇)* ]
Each data point has a probability of .
, (𝑎𝑎, 𝑏𝑏, 0) Class Property
Var[𝑋𝑋] = 𝜎𝜎 - = 𝜇𝜇- 2
𝑝𝑝2 𝑏𝑏
Var[𝑔𝑔(𝑋𝑋)] = E[𝑔𝑔(𝑋𝑋)- ] − E[𝑔𝑔(𝑋𝑋)]- Empirical CDF: = 𝑎𝑎 + , for 𝑛𝑛 = 1, 2, ⋯
𝑝𝑝2", 𝑛𝑛
Covariance: Cov(𝑋𝑋, 𝑌𝑌) = E[𝑋𝑋𝑋𝑋] − E[𝑋𝑋]E[𝑌𝑌] # of observations ≤ 𝑥𝑥
𝐹𝐹2 (𝑥𝑥) =
𝜎𝜎 𝑛𝑛 Note: Included in exam table Appendix B.2.
Coefficient of variation: 𝐶𝐶𝐶𝐶 =
𝜇𝜇 Empirical 100𝑝𝑝() percentile: 𝜋𝜋7 = 𝑥𝑥(27)
𝜇𝜇. 𝜇𝜇/ ∑289, 𝑥𝑥8
Skewness = . ; Kurtosis = / Distributions Relationship
𝜎𝜎 𝜎𝜎 Sample mean: 𝑥𝑥̅ =
𝑛𝑛
Is equivalent to
Moment and Probability Generating Biased sample variance:
Functions ∑289,(𝑥𝑥8 − 𝑥𝑥̅ )- Gamma (1, 𝜃𝜃) Exponential (𝜃𝜃)
Var[𝑋𝑋] =
𝑀𝑀0 (𝑧𝑧) = E[𝑒𝑒 10 ] 𝑛𝑛
Beta (1, 1, 𝜃𝜃) Uniform (0, 𝜃𝜃)
(2) ∑289, 𝑥𝑥8-
𝑀𝑀0 (0) = E[𝑋𝑋 2 ] = − 𝑥𝑥̅ - Binomial (1, 𝑞𝑞) Bernoulli (𝑞𝑞)
𝑛𝑛
(2)
where 𝑀𝑀0 is the 𝑛𝑛() derivative Unbiased sample variance:
Neg. Bin. (1, 𝛽𝛽) Geometric (𝛽𝛽)
𝑃𝑃0 (𝑧𝑧) = E[𝑧𝑧 0 ] ∑2 (𝑥𝑥8 − 𝑥𝑥̅ )-
𝑠𝑠 - = 89,
(2)
𝑃𝑃0 (1) = E[𝑋𝑋(𝑋𝑋 − 1) ⋯ (𝑋𝑋 − 𝑛𝑛 + 1)] 𝑛𝑛 − 1
𝑛𝑛 Sum of Independent Random Variables
where 𝑃𝑃0
(2)
is the 𝑛𝑛() derivative = ⋅ Var[𝑋𝑋]
𝑛𝑛 − 1 𝑿𝑿𝒊𝒊 ∑𝒏𝒏𝒊𝒊9𝟏𝟏 𝑿𝑿𝒊𝒊

Parametric Distributions Gamma (𝛼𝛼8 , 𝜃𝜃) Gamma (∑289, 𝛼𝛼8 , 𝜃𝜃)

Incomplete Gamma Function Normal
Normal â𝜇𝜇8 , 𝜎𝜎8- ä
Γ(𝛼𝛼; 𝑥𝑥) = 1 − Pr(𝑁𝑁 < 𝛼𝛼) â∑289, 𝜇𝜇8 , ∑289, 𝜎𝜎8- ä
where 𝑁𝑁~Poisson(𝜆𝜆 = 𝑥𝑥)
Poisson (𝜆𝜆8 ) Poisson (∑289, 𝜆𝜆8 )

Binomial (𝑚𝑚8 , 𝑞𝑞) Binomial (∑289, 𝑚𝑚8 , 𝑞𝑞)

Neg. Bin. (𝑟𝑟8 , 𝛽𝛽) Neg. Bin. (∑289, 𝑟𝑟8 , 𝛽𝛽)

www.coachingactuaries.com Copyright © 2021 Coaching Actuaries. All Rights Reserved. 1

Mixtures and Splices Franchise deductible: Negative Binomial/Exponential
Bernoulli Shortcut 0, 𝑋𝑋 < 𝑑𝑑 Compound Models
𝑌𝑌 @ = ç
𝑎𝑎, Probability = 𝑞𝑞 𝑋𝑋, 𝑋𝑋 ≥ 𝑑𝑑 𝑁𝑁 ∼ Neg. Binomial (𝑟𝑟, 𝛽𝛽)
If 𝑋𝑋 = ç , then: E[𝑌𝑌 @ ] = E[(𝑋𝑋 − 𝑑𝑑)B ] + 𝑑𝑑 ⋅ 𝑆𝑆(𝑑𝑑) ç ∫
𝑏𝑏, Probability = 1 − 𝑞𝑞 𝑋𝑋 ∼ Exponential (𝜃𝜃)
-
Var[𝑋𝑋] = (𝑎𝑎 − 𝑏𝑏) 𝑞𝑞(1 − 𝑞𝑞) ⇕
Payment per Payment 𝛽𝛽
Special Mixtures 𝑁𝑁 ∼ Binomial ≤𝑟𝑟, ∑
𝑌𝑌 D : payment per payment º 1 + 𝛽𝛽 Ω
If ( 𝑁𝑁 ∣ 𝜆𝜆 ) ∼ Poisson (𝜆𝜆) 𝑋𝑋 ∼ Exponential (𝜃𝜃[1 + 𝛽𝛽])
E[𝑌𝑌 @ ]
where Λ ∼ Gamma (𝛼𝛼, 𝜃𝜃), E[𝑌𝑌 D ] = ; E[𝑌𝑌 @ ] = E[𝑌𝑌 D ] ⋅ 𝑆𝑆(𝑑𝑑)
𝑆𝑆(𝑑𝑑)
then 𝑁𝑁 ∼ Neg. Binomial (𝛼𝛼, 𝜃𝜃). With ordinary deductible 𝑑𝑑, Compound Poisson Models
If ( 𝑋𝑋 ∣ 𝜃𝜃 ) ∼ Normal (𝜃𝜃, 𝑣𝑣) E[𝑌𝑌 D ] = 𝑒𝑒(𝑑𝑑) = E[ 𝑋𝑋 − 𝑑𝑑 ∣ 𝑋𝑋 > 𝑑𝑑 ] A collective risk model where the frequency
where Θ ∼ Normal (𝜇𝜇, 𝑎𝑎), E[(𝑋𝑋 − 𝑑𝑑)B ] follows a Poisson distribution.
=
then 𝑋𝑋 ∼ Normal (𝜇𝜇, 𝑣𝑣 + 𝑎𝑎). 𝑆𝑆(𝑑𝑑)
Discretization – Method of Rounding
If ( 𝑋𝑋 ∣ 𝜃𝜃 ) ∼ Exponential (𝜃𝜃) Let 𝑌𝑌 be the discretized version of 𝑋𝑋, and ℎ
Special Shortcuts for 𝑒𝑒(𝑑𝑑)
where Θ ∼ Inv. Gamma (𝛼𝛼, 𝛽𝛽), be the span.
then 𝑋𝑋 ∼ Pareto (𝛼𝛼, 𝛽𝛽). 𝒆𝒆(𝒅𝒅) Pr(𝑌𝑌 = 𝑦𝑦) = Pr(𝑦𝑦 − 0.5ℎ < 𝑋𝑋 ≤ 𝑦𝑦 + 0.5ℎ)
for 𝑦𝑦 = multiples of ℎ
Frailty Models 𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄𝐄 (𝜽𝜽) 𝜃𝜃
ℎ( 𝑥𝑥 ∣ 𝛬𝛬 ) = 𝛬𝛬 ⋅ 𝑎𝑎(𝑥𝑥) 𝑏𝑏 − 𝑑𝑑 Risk Measures
𝐔𝐔𝐔𝐔𝐔𝐔𝐔𝐔𝐔𝐔𝐔𝐔𝐔𝐔 (𝒂𝒂, 𝒃𝒃)
𝑆𝑆(𝑥𝑥) = 𝑀𝑀? [−𝐴𝐴(𝑥𝑥)], 2 Value-at-Risk (VaR)
!
where 𝐴𝐴(𝑥𝑥) = * 𝑎𝑎(𝑡𝑡) d𝑡𝑡 𝜃𝜃 + 𝑑𝑑 VaR 7 (𝑋𝑋) = 𝐹𝐹0", (𝑝𝑝)
𝐏𝐏𝐏𝐏𝐏𝐏𝐏𝐏𝐏𝐏𝐏𝐏 (𝜶𝜶, 𝜽𝜽)
"# 𝛼𝛼 − 1
Tail-Value-at-Risk (TVaR)
𝑑𝑑
Insurance Applications S-P 𝐏𝐏𝐏𝐏𝐏𝐏𝐏𝐏𝐏𝐏𝐏𝐏 (𝜶𝜶, 𝜽𝜽) TVaR7 (𝑋𝑋) = Ef𝑋𝑋 ∣ 𝑋𝑋 > VaR 7 (𝑋𝑋)g
𝛼𝛼 − 1
𝑌𝑌 @ : payment per loss = VaR7 (𝑋𝑋) + 𝑒𝑒fVaR 7 (𝑋𝑋)g

Policy Limits, 𝑢𝑢 The Ultimate Formula for Insurance

𝑋𝑋, 𝑋𝑋 < 𝑢𝑢 E[𝑌𝑌 @ ] 𝐓𝐓𝐓𝐓𝐓𝐓𝐓𝐓𝒑𝒑 (𝑿𝑿)
𝑌𝑌 @ = 𝑋𝑋 ∧ 𝑢𝑢 = ó
𝑢𝑢, 𝑋𝑋 ≥ 𝑢𝑢 𝑚𝑚 𝑑𝑑
E[(𝑌𝑌 @ )* ]
= E[(𝑋𝑋 ∧ 𝑢𝑢)* ] = 𝛼𝛼(1 + 𝑟𝑟) ≤E ≥𝑋𝑋 ∧ ¥ − E µ𝑋𝑋 ∧ ∂∑ 𝜙𝜙â𝑧𝑧7 ä
1 + 𝑟𝑟 1 + 𝑟𝑟 Normal 𝜇𝜇 + 𝜎𝜎 ƒ ∆
A 1 − 𝑝𝑝
where
= * 𝑥𝑥 * 𝑓𝑓(𝑥𝑥) d𝑥𝑥 + 𝑢𝑢* ⋅ 𝑆𝑆(𝑢𝑢)
' 𝑑𝑑: deductible (set to 0 if not applicable)
A Φâ𝜎𝜎 − 𝑧𝑧7 ä
𝑢𝑢: policy limit (set to ∞ if not applicable) Lognormal E[𝑋𝑋] ⋅ ƒ ∆
= * 𝑘𝑘𝑥𝑥 *", 𝑆𝑆(𝑥𝑥) d𝑥𝑥 1 − 𝑝𝑝
' 𝛼𝛼: coinsurance (set to 1 if not applicable)
E[𝑋𝑋 ∧ 𝑢𝑢] 𝑟𝑟: inflation rate (set to 0 if not applicable)
Increased Limit Factor: 𝐼𝐼𝐼𝐼𝐼𝐼 = Coherence
E[𝑋𝑋 ∧ 𝑏𝑏] 𝑢𝑢
𝑚𝑚: maximum covered loss, i. e. 𝑚𝑚 = + 𝑑𝑑 𝜌𝜌(𝑋𝑋) is coherent if it satisfies the
• 𝑏𝑏: original limit 𝛼𝛼
• 𝑢𝑢: increased limit properties below:
Aggregate Loss Models • Translation invariance: 𝜌𝜌(𝑋𝑋 + 𝑐𝑐) =
Deductibles, 𝑑𝑑 Collective Risk Model 𝜌𝜌(𝑋𝑋) + 𝑐𝑐
Ordinary deductible: If 𝑆𝑆 = ∑E
89, 𝑋𝑋8 for independent 𝑁𝑁 • Positive homogeneity: 𝜌𝜌(𝑐𝑐𝑐𝑐) = 𝑐𝑐 ⋅ 𝜌𝜌(𝑋𝑋)
0, 𝑋𝑋 < 𝑑𝑑 and 𝑋𝑋, then:
𝑌𝑌 @ = (𝑋𝑋 − 𝑑𝑑)B = ç • Subadditivity: 𝜌𝜌(𝑋𝑋 + 𝑌𝑌) ≤ 𝜌𝜌(𝑋𝑋) + 𝜌𝜌(𝑌𝑌)
𝑋𝑋 − 𝑑𝑑, 𝑋𝑋 ≥ 𝑑𝑑
• E[𝑆𝑆] = E[𝑁𝑁]E[𝑋𝑋] • Monotonicity: 𝜌𝜌(𝑋𝑋) ≤ 𝜌𝜌(𝑌𝑌),
E[𝑌𝑌 @ ] = E[(𝑋𝑋 − 𝑑𝑑)B ] = E[𝑋𝑋] − E[𝑋𝑋 ∧ 𝑑𝑑]
• Var[𝑆𝑆] = E[𝑁𝑁]Var[𝑋𝑋] + Var[𝑁𝑁]E[𝑋𝑋]- if Pr(𝑋𝑋 ≤ 𝑌𝑌) = 1
E[(𝑌𝑌 @ )* ] = Ef(𝑋𝑋 − 𝑑𝑑)*B g
# VaR is not coherent because it fails
Impact of Deductibles on Claim Frequency
= * (𝑥𝑥 − 𝑑𝑑)* 𝑓𝑓(𝑥𝑥) d𝑥𝑥 subaddivity.
C For 𝑣𝑣 = Pr(𝑋𝑋 > 𝑑𝑑),
# TVaR is coherent.
= * 𝑘𝑘(𝑥𝑥 − 𝑑𝑑)*", 𝑆𝑆(𝑥𝑥) d𝑥𝑥 𝑵𝑵 𝑵𝑵′
C Tail Weight
E[𝑋𝑋 ∧ 𝑑𝑑] Poisson 𝜆𝜆 𝑣𝑣𝑣𝑣
Loss eliminiation ratio: 𝐿𝐿𝐿𝐿𝐿𝐿 = 1. Fewer positive raw moments ⟹
E[𝑋𝑋] Binomial 𝑚𝑚, 𝑞𝑞 𝑚𝑚, 𝑣𝑣𝑣𝑣 heavier tail
Note: Equivalent to LER formula on the last H!(!) I! (!)
Neg. Binomial 𝑟𝑟, 𝛽𝛽 𝑟𝑟, 𝑣𝑣𝑣𝑣 2. If lim = ∞ or lim = ∞, then
page assuming 𝑏𝑏 = 0. !→# H"(!) !→# I" (!)
numerator has a heavier tail.
3. ℎ(𝑥𝑥) decreases with 𝑥𝑥 ⟹ heavy tail
4. 𝑒𝑒(𝑑𝑑) increases with 𝑑𝑑 ⟹ heavy tail

www.coachingactuaries.com Copyright © 2021 Coaching Actuaries. All Rights Reserved. 2

CONSTRUCTION AND SELECTION OF Choosing from (𝑎𝑎, 𝑏𝑏, 0) Class # of observations < 𝑥𝑥
CONSTRUCTION & SELECTION and 𝐹𝐹2 â𝑥𝑥4" ä =
PARAMETRIC MODELS
OF PARAMETRIC MODELS Two methods to fit data to an (𝑎𝑎, 𝑏𝑏, 0) 𝑛𝑛
class distributions: 𝑝𝑝-𝑝𝑝 Plot
Maximum Likelihood Estimators • Method 1: Compare 𝑥𝑥̅ and 𝑠𝑠 -
Coordinates: ‹𝐹𝐹2 â𝑥𝑥4 ä, 𝐹𝐹 ∗ â𝑥𝑥4 ä›
Steps to Calculating MLE • Method 2: Observe the slope of
*2#

1. 𝐿𝐿(𝜃𝜃) = ∏ 𝑓𝑓(𝑥𝑥) J 2#$! # of observations ≤ 𝑥𝑥

3. 𝑙𝑙+ (𝜃𝜃) = 𝑙𝑙(𝜃𝜃) where 𝐹𝐹2 â𝑥𝑥4 ä =
JK 𝑛𝑛 + 1
2. 𝑙𝑙(𝜃𝜃) = ln 𝐿𝐿(𝜃𝜃) Distribution Method 1 Method 2
4. Set 𝑙𝑙+ (𝜃𝜃) = 0
Poisson 𝑥𝑥̅ = 𝑠𝑠 - 0 Hypothesis Tests
Incomplete Data
𝐻𝐻' : null hypothesis
Binomial 𝑥𝑥̅ > 𝑠𝑠 - Negative
Left-truncated at 𝑑𝑑 𝑓𝑓(𝑥𝑥)⁄𝑆𝑆(𝑑𝑑) 𝐻𝐻, : alternative hypothesis
-
Neg. Binomial 𝑥𝑥̅ < 𝑠𝑠 Positive Reject 𝐻𝐻' when test statistic > critical value
Right-censored at 𝑢𝑢 𝑆𝑆(𝑢𝑢)
Grouped data on Variance of MLE
Pr(𝑎𝑎 < 𝑋𝑋 ≤ 𝑏𝑏) 𝑯𝑯𝟎𝟎 is true 𝑯𝑯𝟎𝟎 is false
interval (𝑎𝑎, 𝑏𝑏] Fisher’s Information
One Parameter:
Type I Correct
Special Cases 𝐼𝐼(𝜃𝜃) = −E0 [𝑙𝑙′′(𝜃𝜃)] Reject 𝑯𝑯𝟎𝟎
Error Decision
Varf𝜃𝜃Œg = [𝐼𝐼(𝜃𝜃)]",
Distribution Shortcuts
Fail to Correct Type II
Gamma, 𝑥𝑥̅ Two Parameters:
reject 𝑯𝑯𝟎𝟎 Decision Error
𝜃𝜃Œ = +
𝑙𝑙L++ (𝛼𝛼, 𝜃𝜃) 𝑙𝑙L,K (𝛼𝛼, 𝜃𝜃)
fixed 𝛼𝛼 𝛼𝛼 𝐼𝐼(𝛼𝛼, 𝜃𝜃) = −E0 ƒ + ∆
𝑙𝑙L,K (𝛼𝛼, 𝜃𝜃) 𝑙𝑙K++ (𝛼𝛼, 𝜃𝜃)
𝜇𝜇̂ = 𝑥𝑥̅ Var[𝛼𝛼–] Covf𝛼𝛼–, 𝜃𝜃Œg Hypothesis Tests: Kolmogorov-Smirnov
[𝐼𝐼(𝛼𝛼, 𝜃𝜃)]", = ’ ÷ Kolmogorov-Smirnov Test
Normal Covf𝛼𝛼–, 𝜃𝜃g Œ Varf𝜃𝜃Œg
∑289,(𝑥𝑥8 − 𝜇𝜇̂ )- Test statistic: 𝐷𝐷 = maxâfl𝐷𝐷â𝑥𝑥4 äfl, fl𝐷𝐷â𝑥𝑥4" äflä
𝜎𝜎– - = QRR 4
𝑛𝑛 Delta Approximation
If data is left-truncated at 𝑑𝑑, then
One-Variable:
∑289, ln 𝑥𝑥8 𝐹𝐹(𝑥𝑥) − 𝐹𝐹(𝑑𝑑)
𝜇𝜇̂ = 𝑑𝑑 - 𝐹𝐹 ∗ (𝑥𝑥) = , for 𝑥𝑥 ≥ 𝑑𝑑
𝑛𝑛 Varf𝑔𝑔â𝜃𝜃Œäg ≈ µ 𝑔𝑔(𝜃𝜃)∂ Varf𝜃𝜃Œg 1 − 𝐹𝐹(𝑑𝑑)
Lognormal 𝑑𝑑𝑑𝑑 If data is right-censored at 𝑢𝑢, then 𝐹𝐹2 (𝑢𝑢) is
∑289,(ln 𝑥𝑥8 − 𝜇𝜇̂ )-
𝜎𝜎– - = Two-Variable: indeterminate.
𝑛𝑛
Varf𝑔𝑔â𝛼𝛼–, 𝜃𝜃Œäg ≈ (𝑔𝑔L+ )- Var[𝛼𝛼–]
Kolmogorov-Smirnov Test Properties
Poisson 𝜆𝜆— = 𝑥𝑥̅ + 2𝑔𝑔L+ 𝑔𝑔K+ Covf𝛼𝛼–, 𝜃𝜃Œ g • Individual data only
Binomial, 𝑥𝑥̅ + (𝑔𝑔K+ )- Varf𝜃𝜃Œg • Continuous fit only
𝑞𝑞– =
fixed 𝑚𝑚 𝑚𝑚 • Lower critical value for censored data
Confidence Interval
Neg. Binomial, 𝑥𝑥̅ • If parameters are estimated, critical value
𝛽𝛽— = ⁄ f𝜃𝜃Œg
𝜃𝜃Œ ± 𝑧𝑧(,B7)/- ŸVar should be adjusted
fixed 𝑟𝑟 𝑟𝑟
• Lower critical value if sample size is large
Graphical Analyses • No discretion
Zero-Truncated Distribution: 𝐷𝐷(𝑥𝑥) Plot • Uniform weight on all parts
• Match E[𝑋𝑋 : ] to 𝑥𝑥̅ Graph the difference between empirical of distribution
Zero-Modified Distribution: CDF and fitted CDF
• Match 𝑝𝑝'; to the proportion of
zero observations
• Match E[𝑋𝑋 ; ] to 𝑥𝑥̅
Uniform Distribution on (0, 𝜃𝜃):
• 𝜃𝜃Œ = max(𝑥𝑥, , 𝑥𝑥- , … , 𝑥𝑥2 )

Peak: 𝐷𝐷â𝑥𝑥4 ä = 𝐹𝐹2 â𝑥𝑥4 ä − 𝐹𝐹∗ â𝑥𝑥4 ä

Valley: 𝐷𝐷â𝑥𝑥4" ä = 𝐹𝐹2 â𝑥𝑥4" ä − 𝐹𝐹∗ â𝑥𝑥4 ä
# of observations ≤ 𝑥𝑥
where 𝐹𝐹2 â𝑥𝑥4 ä =
𝑛𝑛

www.coachingactuaries.com Copyright © 2021 Coaching Actuaries. All Rights Reserved. 3

Hypothesis Tests: Chi-Square CREDIBILITY Predictive Distribution
CREDIBILITY
Goodness-of-Fit Revised unconditional distribution
Chi-Square Goodness-of-Fit Test Classical Credibility (w.r.t. parameter) of the model
* - a.k.a. Limited Fluctuation Credibility Predictive density function: 𝑓𝑓(𝑥𝑥 ∣ data)
â𝐸𝐸4 − 𝑂𝑂4 ä
Test statistic: 𝜒𝜒 - = · Predictive Mean = Bayesian Premium
𝐸𝐸4 Full Credibility
49,
*
𝑂𝑂4- # of exposures needed for full credibility, 𝑛𝑛T : Loss Function
=· − 𝑛𝑛 Full credibility of aggregate claims:
𝐸𝐸4
49, 𝑧𝑧(,B7)⁄- - Function Posterior
where 𝑛𝑛T = ≥ ¥ (𝐶𝐶𝑉𝑉H- )
𝑘𝑘 to Î, 𝜽𝜽ä
𝒍𝒍â𝜽𝜽 Estimation,
• 𝑘𝑘: # of groups Minimize Î
𝜽𝜽
# of claims needed for full credibility, 𝑛𝑛V :
• 𝐸𝐸4 : expected # of observations in group 𝑗𝑗
Full credibility of aggregate claims:
• 𝑂𝑂4 : actual # of observations in group 𝑗𝑗 Squared- - Posterior
𝑧𝑧(,B7)⁄- - 𝜎𝜎E- â𝜃𝜃Œ − 𝜃𝜃ä
Degrees of freedom = 𝑘𝑘 − 1 − 𝑟𝑟 where 𝑛𝑛V = ≥ ¥ ’ + 𝐶𝐶𝑉𝑉0- ÷ error loss mean
𝑘𝑘 𝜇𝜇E
• 𝑟𝑟: # of estimated parameters
• Full credibility of claim frequency: set Absolute Posterior
fl𝜃𝜃Œ − 𝜃𝜃fl
Chi-Square Goodness-of-Fit Test Properties 𝐶𝐶𝑉𝑉0- = 0 loss median
• Individual and grouped data • Full credibility of claim severity:
• Continuous and discrete fit 𝜎𝜎E- Zero-one 1 for 𝜃𝜃Œ ≠ 𝜃𝜃 Posterior
set =0 loss 0 for 𝜃𝜃Œ = 𝜃𝜃 mode
• No adjustments to critical value for 𝜇𝜇E
𝑛𝑛V
censored data 𝑛𝑛V = 𝑛𝑛T ⋅ 𝜇𝜇E ⟺ 𝑛𝑛T =
𝜇𝜇E Bühlmann Credibility
• If parameters are estimated, critical
Expected Hypothetical Mean (EHM):
value is automatically adjusted via Partial Credibility
𝜇𝜇 = EfE[𝑋𝑋 ∣ 𝜃𝜃]g
degrees of freedom Credibility premium: 𝑃𝑃W = 𝑍𝑍𝑥𝑥̅ + (1 − 𝑍𝑍)𝑀𝑀
Expected Process Variance (EPV):
• No change for critical value if sample = 𝑀𝑀 + 𝑍𝑍(𝑥𝑥̅ − 𝑀𝑀)
size is large 𝑣𝑣 = EfVar[𝑋𝑋 ∣ 𝜃𝜃]g
where
• Data needs to be grouped according to 𝐸𝐸4 • 𝑀𝑀: manual premium Variance of Hypothetical Mean (VHM):

• More weights on intervals with poor fit • 𝑍𝑍: credibility factor/credibility 𝑎𝑎 = VarfE[𝑋𝑋 ∣ 𝜃𝜃]g
𝑣𝑣
𝑛𝑛 𝑛𝑛′ Bühlmann 𝑘𝑘: 𝑘𝑘 =
𝑎𝑎
Hypothesis Tests: Likelihood Ratio Square Root Rule: 𝑍𝑍 = Ë = Ë 𝑛𝑛
𝑛𝑛T 𝑛𝑛V Bühlmann Credibility Factor: 𝑍𝑍 =
Test statistic: 𝑇𝑇 = 2[𝑙𝑙(𝜃𝜃, ) − 𝑙𝑙(𝜃𝜃' )] 𝑛𝑛 + 𝑘𝑘
Degrees of freedom where
• 𝑛𝑛: actual # of exposures Bühlmann Credibility Premium:
= # of free parameters in 𝐻𝐻,
• 𝑛𝑛′: actual # of claims 𝑃𝑃W = 𝑍𝑍𝑥𝑥̅ + (1 − 𝑍𝑍)𝜇𝜇
− # of free parameters in 𝐻𝐻'
= 𝜇𝜇 + 𝑍𝑍(𝑥𝑥̅ − 𝜇𝜇)
Bayesian Credibility Note: 𝑍𝑍 and 𝑥𝑥̅ have the same 𝑛𝑛.
Score-Based Approaches
Two types of criteria: Model Distribution
Distribution of model conditioned on a Bühlmann As Least Squares
• Schwarz Bayesian Criterion (SBC), a.k.a.
parameter Estimate of Bayesian
Bayesian Information Criterion (BIC) -
• Akaike Information Criterion (AIC) Model density function: 𝑓𝑓( 𝑥𝑥 ∣ 𝜃𝜃 ) Minimize · ≥𝑝𝑝! â𝑌𝑌! − 𝑌𝑌Œ! ä ¥ where
QRR !
Prior Distribution 𝑌𝑌! : Bayesian estimate given 𝑋𝑋, = 𝑥𝑥
𝑟𝑟
SBC/BIC 𝑙𝑙 − ln 𝑛𝑛 Initial distribution of the parameter 𝑌𝑌Œ! : Bühlmann estimate given 𝑋𝑋, = 𝑥𝑥
2
Prior density function: 𝜋𝜋(𝜃𝜃)
AIC 𝑙𝑙 − 𝑟𝑟 Properties of a Bayesian/Bühlmann graph
Posterior Distribution • Bühlmann estimates are on a straight line
where Revised distribution of the parameter • Bayesian estimates are within the range
𝑙𝑙: log-likelihood Posterior density function: 𝜋𝜋(𝜃𝜃 ∣ data) of hypothetical means
𝑟𝑟: # of estimated parameters 𝑓𝑓( data ∣ 𝜃𝜃 ) ⋅ 𝜋𝜋(𝜃𝜃)
𝜋𝜋(𝜃𝜃 ∣ data) = # • There are Bayesian estimates above and
𝑛𝑛: sample size ∫"# 𝑓𝑓( data ∣ 𝜃𝜃 ) ⋅ 𝜋𝜋(𝜃𝜃) d𝜃𝜃 below the Bühlmann line
Select model with the highest SBC Note: Use numerator and domain to check if • Bühlmann estimates are between the
or AIC value. (𝜃𝜃 ∣ data) follows a distribution in the exam sample mean and theoretical mean
tables to skip integration.

www.coachingactuaries.com Copyright © 2021 Coaching Actuaries. All Rights Reserved. 4

Conjugate Priors Uniform/S-P Pareto Non-uniform Exposures
2
Poisson/Gamma • Model: ( 𝑋𝑋 ∣ 𝜃𝜃 ) ∼ Uniform (0, 𝜃𝜃) ∑X89, ∑49,
%
𝑚𝑚84 𝑥𝑥84
𝜇𝜇̂ = = 𝑥𝑥̅
• Model: ( 𝑋𝑋 ∣ 𝜆𝜆 ) ∼ Poisson (𝜆𝜆) • Prior: 𝜃𝜃 ∼ S-P Pareto (𝛼𝛼, 𝛽𝛽) 𝑚𝑚
• Prior: 𝜆𝜆 ∼ Gamma (𝛼𝛼, 𝜃𝜃) 2 -
Posterior ∑X89, ∑49,
%
𝑚𝑚84 â𝑥𝑥84 − 𝑥𝑥̅ 8 ä
Posterior ( 𝜃𝜃 ∣ data ) ∼ S-P Pareto (𝛼𝛼 ∗ , 𝛽𝛽∗ ) 𝑣𝑣– =
∑X89,(𝑛𝑛8 − 1)
( 𝜆𝜆 ∣ data ) ∼ Gamma (𝛼𝛼 ∗ , 𝜃𝜃 ∗ ) • 𝛼𝛼 ∗ = 𝛼𝛼 + 𝑛𝑛
• 𝛼𝛼 = 𝛼𝛼 + ∑289, 𝑥𝑥8
∗
• 𝛽𝛽∗ = max(𝛽𝛽, 𝑥𝑥, , … , 𝑥𝑥2 ) ∑X89, 𝑚𝑚8 (𝑥𝑥̅8 − 𝑥𝑥̅ )- − 𝑣𝑣–(𝑟𝑟 − 1)
𝑎𝑎– =
", 𝑚𝑚 − 𝑚𝑚", ∑X89, 𝑚𝑚8-
,
• 𝜃𝜃 ∗ = ‹ + 𝑛𝑛› Predictive
K
−
Predictive Credibility premium for a risk in Class 𝑖𝑖:
( 𝑋𝑋 ∣ data ) ∼ Neg. Bin. (𝑟𝑟 = 𝛼𝛼 ∗ , 𝛽𝛽 = 𝜃𝜃 ∗ ) 𝑃𝑃ŒW = 𝑍𝑍—8 𝑥𝑥̅8 + â1 − 𝑍𝑍—8 ä𝜇𝜇̂
Exact Credibility
Bayesian estimate = Bühlmann estimate Note: 𝑍𝑍—8 and 𝑥𝑥̅8 have the same 𝑛𝑛.
Binomial/Beta
• Poisson/Gamma Balancing the Estimators
• Model: ( 𝑋𝑋 ∣ 𝑞𝑞 ) ∼ Binomial (𝑚𝑚, 𝑞𝑞)
• Binomial/Beta ∑X89, 𝑍𝑍8 𝑥𝑥̅ 8
• Prior: 𝑞𝑞 ∼ Beta (𝑎𝑎, 𝑏𝑏, 1) Estimate EHM as: 𝜇𝜇̂ = X
• Exponential/Inv. Gamma ∑89, 𝑍𝑍8
Posterior • Normal/Normal
( 𝑞𝑞 ∣ data ) ∼ Beta (𝑎𝑎∗ , 𝑏𝑏∗ , 1) Empirical Bayes
• 𝑎𝑎∗ = 𝑎𝑎 + ∑289, 𝑥𝑥8 Empirical Bayes Semi-Parametric Methods
• 𝑏𝑏∗ = 𝑏𝑏 + [𝑛𝑛(𝑚𝑚) − ∑289, 𝑥𝑥8 ] Non-Parametric Methods To estimate 𝑣𝑣:
Predictive Uniform Exposures
− ∑X89, ∑249, 𝑥𝑥84 Model 𝒗𝒗
Ò
𝜇𝜇̂ = = 𝑥𝑥̅
𝑟𝑟 ⋅ 𝑛𝑛
Poisson 𝑥𝑥̅
Exponential/Inv. Gamma -
∑X89, ∑249,â𝑥𝑥84 − 𝑥𝑥̅8 ä Neg. Binomial with
• Model: ( 𝑋𝑋 ∣ 𝜃𝜃 ) ∼ Exponential (𝜃𝜃) 𝑣𝑣– = 𝑥𝑥̅ (1 + 𝛽𝛽)
𝑟𝑟(𝑛𝑛 − 1) fixed 𝛽𝛽
• Prior: 𝜃𝜃 ∼ Inv. Gamma (𝛼𝛼, 𝛽𝛽)
Gamma with fixed 𝜃𝜃 𝑥𝑥̅ 𝜃𝜃
∑X89,(𝑥𝑥̅ 8 − 𝑥𝑥̅ )- 𝑣𝑣–
Posterior 𝑎𝑎– = −
𝑟𝑟 − 1 𝑛𝑛 Binomial with fixed 𝑞𝑞 𝑥𝑥̅ (1 − 𝑞𝑞)
( 𝜃𝜃 ∣ data ) ∼ Inv. Gamma (𝛼𝛼 ∗ , 𝛽𝛽∗ )
∗
• 𝛼𝛼 = 𝛼𝛼 + 𝑛𝑛 To estimate 𝜇𝜇 and 𝑎𝑎, use the non-parametric
• 𝛽𝛽∗ = 𝛽𝛽 + ∑289, 𝑥𝑥8 method formulas shown above.
Predictive
( 𝑋𝑋 ∣ data ) ∼ Pareto (𝛼𝛼 = 𝛼𝛼 ∗ , 𝜃𝜃 = 𝛽𝛽∗ )

Normal/Normal
• Model: ( 𝑋𝑋 ∣ 𝜃𝜃 ) ∼ Normal (𝜃𝜃, 𝑣𝑣)
• Prior: 𝜃𝜃 ∼ Normal (𝜇𝜇, 𝑎𝑎)

Posterior
( 𝜃𝜃 ∣ data ) ∼ Normal (𝜇𝜇∗ , 𝑎𝑎∗ )
∗
• 𝜇𝜇 = 𝑍𝑍𝑥𝑥̅ + (1 − 𝑍𝑍)𝜇𝜇
• 𝑎𝑎∗ = (1 − 𝑍𝑍)𝑎𝑎
Predictive
𝜇𝜇 = 𝜇𝜇∗ ,
( 𝑋𝑋 ∣ data ) ∼ Normal ≤ - ∑
𝜎𝜎 = 𝑣𝑣 + 𝑎𝑎∗

www.coachingactuaries.com Copyright © 2021 Coaching Actuaries. All Rights Reserved. 5

SHORT-TERM
SHORT-TERM
SHORT-TERM
INSURANCES
INSURANCES
INSURANCES
SHORT-TERM INSURANCES Liability
Liability
Liability Affordable
Affordable
AffordableCare
Care
Care
Act
Act
Act
(ACA)
(ACA)
(ACA)
Low
Low
Low frequency,
frequency,
frequency, high
high
high severity
severity
severityclaims
claims
claims with
with
with •••Individual
Individual
Individual
mandate
mandate
mandate
Insurance
Insurance
Insurance Coverages
Coverages
Coverages potentially
potentially
potentially high
high
high litigation
litigation
litigation costs
costs
costs (must
(must
(must
have
have
haveinsurance)
insurance)
insurance)
Automobile
Automobile
Automobile •••Claims-made
Claims-made
Claims-made basis:
basis:
basis: losses
losses
lossescovered
covered
covered only
only
only •••Partially
Partially
Partially
subsidized
subsidized
subsidized
insurance
insurance
insurance
•••Third-party
Third-party
Third-partybodily
bodilyinjury
bodily injury and
injuryand
and ififreported
ifreported
reported during
during
during policy
policy
policyperiod;
period;
period; insurer
insurer
insurer •••Plans
Plans
Plans
must
must
must
cover
cover
cover
10
10
10EHBs
EHBs
EHBs
property
property
propertydamage
damage
damage may may
may offer
offer
tail
offer tail
coverage
tail coverage
coverage •••No
No
No
cost
cost
cost
sharing
sharing
sharing for
for
for
preventive
preventive
preventive services
services
services
•••Medical
Medical
Medicalpayments
payments
payments (tort);
(tort); Personal
(tort);Personal
Personal •••Occurrence:
Occurrence:
Occurrence:losses
losses
losses covered
covered
coveredeven
even
evenififif •••Plans
Plans
Plans
must
must
must
meet
meet
meet AV
AVAV
metal
metal
metal
level
level
level
injury
injuryprotection
injuryprotection(no-fault)
protection(no-fault)
(no-fault) reported
reported
reportedafter
after
afterpolicy
policy
policyperiod
period
period •••Plans
Plans
Plans
must
must
must
set
set
setout-of-pocket
out-of-pocket
out-of-pocketlimit
limit
limit
below
below
below
•••Unidentified,
Unidentified,underinsured,
Unidentified,underinsured,
underinsured, uninsured
uninsured
uninsured government
government
government limit
limit
limit
Major
Major
MajorMedical
Medical
Medical
motorists
motorists
motoristsprotection
protection
protection •••Plans
Plans
Planssold
sold
sold
on
on
on
public
public
public
exchanges
exchanges
exchanges
must
must
must
Provider
Provider
Provider networks
networks
networks
•••Collision
Collisionand
Collisionandother
andother
otherthan
than collision
thancollision
collision satisfy
satisfy
satisfy
certain
certain
certain
tests/standards
tests/standards
tests/standards
•••Build
Build
Buildown:
own:
own:common
common
common for
for
forcarriers
carriers
carrierswith
with
with
Homeowners
Homeowners
Homeowners geographically
geographically
geographicallyconcentrated
concentrated
concentrated insureds
insureds
insureds Dental
Dental
Dental
Doctrine
Doctrine
Doctrine ofofof
proximate
proximatecause
proximatecause
cause •••Contract
Contract
Contractwith
with
with networks
networks
networks for
for
forhire:
hire:
hire: Usually
Usually
Usually group
group
group policies
policies
policies because:
because:
because:
•••AAloss
Aloss isisis
loss only
onlycovered
onlycoveredififaifacovered
covered acoveredperil
coveredperil
peril common
common
common forfor
for carriers
carriers
carrierswith
with
withgeographically
geographically
geographically •••High
High
Highfrequency,
frequency,
frequency, low
lowlowseverity
severity
severity
isisis
the
the proximate
theproximate
proximate cause
cause
causeofofof diverse
diverse
diverseinsureds
insureds
insureds •••Tax
Tax
Taxsubsidy
subsidy
subsidy
aacovered
acovered
covered consequence.
consequence.
consequence.
Main
Main
Maintypes
types
types
ofofof
health
health
health
insurance
insurance
insurance
plans:
plans:
plans: Product
Product
Product design
design
design drivers
drivers
drivers
Coinsurance
Coinsurance
Coinsurance Provision:
Provision:
Provision: •••Health
Health
Health
Maintenance
Maintenance
Maintenance Organization
Organization
Organization
(HMO)
(HMO)
(HMO) •••Induced
Induced
Induced utilization:
utilization:
utilization:over-utilize
over-utilize
over-utilize benefits
benefits
benefits
Claim
Claim
Claim Payment,
Payment,
Payment, 𝑃𝑃𝑃𝑃𝑃𝑃 •••Preferred
Preferred
Preferred
Provider
Provider
Provider
Organization
Organization
Organization
(PPO)
(PPO)
(PPO) because
because
because insurance
insurance
insurance coverage
coverage
coverage isisis
present
present
present
𝐼𝐼 𝐼𝐼 𝐼𝐼 •••Accumulated
Accumulated
Accumulated untreated
untreated
untreatedconditions:
conditions:
conditions:
minmin
min 0𝐼𝐼, ⋅ ⋅𝐿𝐿6
0𝐼𝐼,
0𝐼𝐼, ⋅𝐿𝐿6
𝐿𝐿6
, , , 𝐼𝐼 𝐼𝐼<
𝐼𝐼<<
𝑐𝑐𝑐𝑐
𝑐𝑐𝑐𝑐
𝑐𝑐𝑐𝑐 Insurance
Insurance
Insurance
benefit
benefit
benefit
calculations
calculations
calculations
===
/// 𝑐𝑐𝑐𝑐 𝑐𝑐𝑐𝑐
𝑐𝑐𝑐𝑐 ability
ability
ability
tototo
postpone
postpone
postpone treatment
treatment
treatment until
until
until
min(𝐼𝐼,
min(𝐼𝐼,
min(𝐼𝐼, 𝐿𝐿)𝐿𝐿)𝐿𝐿)
, , , 𝐼𝐼 𝐼𝐼≥ 𝐼𝐼≥≥
𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐
𝑐𝑐𝑐𝑐 •••Provider
Provider
Provider •••Out-of-pocket
Out-of-pocket
Out-of-pocket limits
limits
limits
insurance
insurance
insurance isisis
present
present
present
discounts •••Maximum
discounts
discounts Maximum
Maximum limits
limits
limits
Workers
Workers
Workers Compensation
Compensation
Compensation •••Anti-selection:
Anti-selection:
Anti-selection: Individuals
Individuals
Individualschoose
choose
choose dental
dental
dental
•••Deductibles
Deductibles •••Internal
Deductibles Internal
Internal
limits
limits
limits
Prior
Prior tototo
Prior the
thepassage
the passage
passage ofofof
workers
workers
workers comp
complaws,
complaws,
laws, coverage
coverage
coverage that
that
thatbenefits
benefits
benefits
them
them
them
thethe
the
most
most
most
•••Coinsurance
Coinsurance •••Copays
Coinsurance Copays
Copays
ititwas
itwasdifficult
was difficult
difficultfor
for workers
forworkers
workers tototo
get
get
get •••ACA:
ACA:
ACA:pediatric
pediatric
pediatric dental
dental
dental
coverage
coverage
coverage required
required
required
compensation
compensation
compensation for
for injury/illness.
forinjury/illness.
injury/illness. Reasons:
Reasons:
Reasons: Related
Related
Relatedproducts
products
products
•••Doctrine
Doctrine
Doctrine ofofof
contributory
contributory
contributory negligence
negligence
negligence Limits
Limits
Limitstototo Coverage
Coverage
Coverage
•••Comprehensive
Comprehensive
Comprehensive major
major
major medical:
medical:
medical:
covers
covers
covers
Disappearing
Disappearing
Disappearing deductible:
deductible:
deductible: deductible
deductible
deductible
•••Fellow-servant
Fellow-servant
Fellow-servant doctrine
doctrine
doctrine small
small
smallfrequent
frequent
frequent
and
and
andlarge
large
large
infrequent
infrequent
infrequent
decreases
decreases
decreases linearly
linearly
linearly over
over over aaspecific
aspecific
specific range
range
range
•••Assumption-of-risk
Assumption-of-risk
Assumption-of-risk doctrine
doctrine
doctrine costs;
costs;
costs;
small
small
small
deductibles
deductibles
deductibles
𝑑𝑑,𝑑𝑑,𝑑𝑑, 𝑋𝑋𝑋𝑋𝑋𝑋
≤≤≤ 𝑎𝑎𝑎𝑎𝑎𝑎
•••Catastrophic
Catastrophic
Catastrophic major
major
major medical:
medical:
medical:
covers
covers
covers 𝑏𝑏𝑏𝑏−
𝑏𝑏−− 𝑋𝑋𝑋𝑋𝑋𝑋
Workers
Workers
Workers can
can
can
expect:
expect:
expect: 𝐷𝐷𝐷𝐷𝐷𝐷
===<<𝑑𝑑<𝑑𝑑0𝑑𝑑00 66,6, , 𝑎𝑎𝑎𝑎<
𝑎𝑎<<𝑋𝑋𝑋𝑋𝑋𝑋
≤≤≤ 𝑏𝑏𝑏𝑏𝑏𝑏
expenses
expenses
expenses higher
higher
higher
than
than
than regular
regular
regular
major
major
major 𝑏𝑏𝑏𝑏−
𝑏𝑏−− 𝑎𝑎𝑎𝑎𝑎𝑎
•••Medical
Medical
Medicalcare
care
care
benefits
benefits
benefits
medical
medical
medical expenses;
expenses;
expenses;very
very
veryhigh
high
high
deductibles
deductibles
deductibles 0,0,0, 𝑋𝑋𝑋𝑋𝑋𝑋
>>> 𝑏𝑏𝑏𝑏𝑏𝑏
•••Disability
Disability
Disability
income
income
incomebenefits
benefits
benefits
•••Short-term
Short-term
Short-term medical:
medical:
medical: single
single
single
limited
limited
limitedterm;
term;
term;
•••Death
Death
Death
benefits
benefits
benefits Claim
Claim
Claimpayment:
payment:
payment:
simpler
simpler
simplerunderwriting;
underwriting;
underwriting; lower
lower
lower insurer
insurer
insurer
cost
cost
cost 0,0,0, 𝑋𝑋𝑋𝑋𝑋𝑋
<<< 𝑑𝑑𝑑𝑑𝑑𝑑
•••Rehabilitation
Rehabilitation
Rehabilitation services
services
servicesand
and
and
benefits
benefits
benefits ⎧⎧⎧
•••High
High
Highrisk
risk
risk
pool
pool
pool
plan:
plan:
plan:insures
insures
insurespeople
people
people with
with
with ⎪⎪⎪ 𝑋𝑋𝑋𝑋− 𝑋𝑋−− 𝑑𝑑,𝑑𝑑,𝑑𝑑, 𝑑𝑑𝑑𝑑≤
𝑑𝑑≤≤ 𝑋𝑋𝑋𝑋𝑋𝑋
≤≤≤ 𝑎𝑎𝑎𝑎𝑎𝑎
Fire
Fire
Fire no
no
noinsurance
insurance
insurance due
due
duetototo 𝑌𝑌𝑌𝑌𝑌𝑌
=== 𝑏𝑏𝑏𝑏−𝑏𝑏−−𝑋𝑋𝑋𝑋𝑋𝑋
𝑋𝑋𝑋𝑋−
⎨⎨⎨ 𝑑𝑑𝑑𝑑0𝑑𝑑00
𝑋𝑋−− 66, 6, , 𝑎𝑎𝑎𝑎<
𝑎𝑎<< 𝑋𝑋𝑋𝑋𝑋𝑋
≤≤≤ 𝑏𝑏𝑏𝑏𝑏𝑏
Standard
Standard
Standard fire
fire policy
firepolicy
policy (SFP)
(SFP)
(SFP)covers
coversdirect
covers direct
direct pre-existing
pre-existing
pre-existing condition
condition
condition ⎪⎪⎪ 𝑏𝑏𝑏𝑏−𝑏𝑏−−𝑎𝑎𝑎𝑎𝑎𝑎
⎩⎩⎩ 𝑋𝑋,𝑋𝑋,𝑋𝑋, 𝑋𝑋𝑋𝑋𝑋𝑋
>>> 𝑏𝑏𝑏𝑏𝑏𝑏
loss
lossfrom
loss from
fromfire
fire and
fireand
andlightning,
lightning,
lightning,and
andatatat
and least
least
least •••Consumer
Consumer
Consumer directed
directed
directed plan:
plan:
plan:high-deductible
high-deductible
high-deductible
one
one
oneofofof
the
thefollowing:
the following:
following: plan
plan
plan++personal
+personal
personal spending
spending
spending account
account
account
•••Personal
Personal
Personal coverage
coverage
coverage oooHSA
HSA
HSAcan
can
canonly
only
only
be
bebe
opened
opened
opened with
with
with aaHDHP
aHDHP
HDHP
•••Commercial
Commercial
Commercial coverage
coverage
coverage oooTax-free
Tax-free
Tax-freecontributions,
contributions,
contributions, earnings,
earnings,
earnings, and
and
and
•••Increased
Increased
Increased covered
covered
covered perils
perils
perils withdrawals
withdrawals
withdrawals (for
(for
(formedical
medical
medical purposes)
purposes)
purposes)
•••Increased
Increased
Increased covered
covered
covered loss
loss
loss

Marine
Marine
Marine
Inland
Inland
Inlandmarine
marineinsurance
marineinsurance
insurance covers:
covers:
covers:
•••Domestic
Domestic
Domesticshipments
shipments
shipments
•••Instrumentalities
Instrumentalitiesofofof
Instrumentalities transportation
transportation
transportationand
and
and
communication
communication
communication (bridges,
(bridges,
(bridges, tunnels,
tunnels,etc.)
etc.)
tunnels, etc.)
•••Personal
Personalproperty
Personalproperty
propertyfloater
floaterrisks
risks
floater risks

www.coachingactuaries.com Copyright © 2021 Coaching Actuaries. All Rights Reserved. 6

Loss Reserving Data Preparation
Expected Loss Ratio Method Losses
1. 𝐿𝐿J!"#. = 𝑃𝑃% ⋅ 𝐸𝐸𝐸𝐸𝐸𝐸
Losses
2. 𝑅𝑅 = 𝐿𝐿J!"#. − 𝐿𝐿&

Chain-Ladder Method Aggregation Develop to Trending

a.k.a. Loss Development Triangle Method Ultimate
• Calendar Year (CY) • Trend Period
1. 𝑓𝑓'!"#. = ∏)
(*+,- 𝑓𝑓( • Accident Year (AY) • Loss Development • Trend Factor
• Policy Year (PY) Factors
2. 𝐿𝐿J!"#.
' = 𝐿𝐿',+ ⋅ 𝑓𝑓'!"#.

3. 𝑅𝑅 = 𝐿𝐿J!"#. − 𝐿𝐿& Projected Losses

Bornhuetter-Ferguson Method Incurred losses for CY 𝑖𝑖: 𝐿𝐿8' = 𝐿𝐿&' + 𝑅𝑅' − 𝑅𝑅'5-
1
𝑅𝑅 = 𝑃𝑃% ⋅ 𝐸𝐸𝐸𝐸𝐸𝐸 ⋅ 01 − !"#. 6 where 𝑅𝑅' is the reserves at the end of CY 𝑖𝑖
𝑓𝑓
where 𝑓𝑓 !"#. is calculated based on the chain-ladder method Incurred losses for AY or PY 𝑖𝑖: 𝐿𝐿8' = 𝐿𝐿&' + 𝑅𝑅'
where 𝑅𝑅' is the reserves as of the valuation date
Alternatively,
1 Expenses and Profit
𝑅𝑅 = 𝑤𝑤 ⋅ 𝑅𝑅/0 + (1 − 𝑤𝑤) ⋅ 𝑅𝑅01 where 𝑤𝑤 =
𝑓𝑓 !"#. 𝐸𝐸9
Variable Expense Ratio: 𝑉𝑉 =
𝑃𝑃
Frequency-Severity Method 𝐸𝐸:
Alternate Method: Fixed Expense Ratio: 𝐹𝐹 =
𝑃𝑃
1. Apply the chain-ladder method to frequency and severity Permissible Loss Ratio: PLR = 1 − 𝑉𝑉 − 𝑄𝑄; ,
separately where 𝑄𝑄; is the target profit and contingencies ratio
2. 𝐿𝐿J!"#. = 𝑁𝑁
X !"#. ⋅ 𝑋𝑋J !"#.
Premium
3. 𝑅𝑅 = 𝐿𝐿J!"#. − 𝐿𝐿&

Closure Method: Premium

Frequency
2
1. 𝑐𝑐',( = 4$%&. !,#
3 53! !,#() Aggregation Current Rate Level
X'!"#. − 𝑁𝑁',(5- ]
2. 𝑛𝑛Z',( = 𝑐𝑐̂( \𝑁𝑁 • Calendar Year (CY) • Extension of
Aggregate • Policy Year (PY) Exposures Method
• Parallelogram Method
1. 𝑙𝑙_',( = 𝑛𝑛Z',( ⋅ 𝑥𝑥Z',(
2. 𝑅𝑅 = ∑',(67 𝑙𝑙_',( , where 𝑦𝑦 is the valuation CY
Premium at Current Rates

Unearned premium for CY 𝑖𝑖: 𝑃𝑃'< = 𝑃𝑃'= − 𝑃𝑃'% + 𝑃𝑃'5-

<

Extension of Exposures Method

Recalculates the premiums of historical policies under the current
rate level

Parallelogram Method
Calculates average factors to be applied to the aggregate historical
premiums to make them on-level

www.coachingactuaries.com Copyright © 2021 Coaching Actuaries. All Rights Reserved. 7

Ratemaking
Ratemaking Other
Other Topics
Topics
Loss
LossRatio
RatioMethod
Method Individual
IndividualRisk
RiskRating
Rating Plans
Plans
𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿
++ 𝐹𝐹 𝐹𝐹 • •Prospective
Prospective
Indicated
Indicated
Avg.
Avg.
Rate
Rate
Change
Change
== −−11
1−
1− 𝑉𝑉 𝑉𝑉
−− 𝑄𝑄;𝑄𝑄; Experience
o o Experience rating
rating plans
plans letlet
thethe insurer
insurer modify
modify anan insured’s
insured’s
𝐿𝐿𝑅𝑅𝐿𝐿𝑅𝑅 premium
premium based
based onon their
their actual
actual loss
loss experience.
experience.
' '
Indicated
Indicated
Relativity
Relativity
' =
' =
Current
Current
Relativity
Relativity
' ⋅' ⋅ actual
actual experience
experience
𝐿𝐿𝑅𝑅𝐿𝐿𝑅𝑅
>?@A>?@A 𝑀𝑀𝑀𝑀== 𝑍𝑍 𝑍𝑍
00 6+6+ (1(1
−− 𝑍𝑍)𝑍𝑍)
expected
expected experience
experience
Indicated
Indicated
Base
Base
Rate
Rate
==Current
CurrentBase
BaseRate
Rate Schedule
o o Schedule rating
rating plans
plans letlet
thethe insurer
insurer modify
modify anan insured’s
insured’s
11 ++ Indicated
Indicated Avg.
Avg. Rate
Rate Change
Change premium
premium based
based onon thethe risk
risk characteristics
characteristics that
that have
have not
not been
been
⋅ ⋅
Off-Balance
Off-Balance Factor
Factor reflected
reflected ininthethe past
past loss
loss experience.
experience.
Indicated
Indicated Avg.
Avg. Relativity
Relativity Composite
o o Composite rating
rating plans
plans letlet
thethe insurer
insurer rate
rate a policy
a policy with
with
Off-Balance
Off-BalanceFactor
Factor== multiple
multiple coverages
coverages using
using a single
a single exposure
exposure base,
base, instead
instead ofof
aa
Current Avg.
Current Relativity
Avg. Relativity
separate
separate exposure
exposure base
base forfor each
each coverage.
coverage.
Pure
Pure
Premium
Premium Method
Method Large
o o Large deductible
deductible policies
policies allow
allow thethe insureds
insureds toto self-insure
self-insure upuptoto
𝐿𝐿z + 𝐸𝐸z:𝐸𝐸z:
𝐿𝐿z + the
the deductible
deductible amount.
amount.
Indicated
Indicated
Avg.
Avg.
Rate
Rate
==
1−
1− 𝑉𝑉 𝑉𝑉
−− 𝑄𝑄;𝑄𝑄;
Retrospective
• •Retrospective rating
rating plans
plans determine
determine a policy's
a policy's premium
premium using
using thethe
Avg.
Avg. Rate
Rate' ' actual
actual loss
loss experience
experience ofof the
the same
same policy
policy period.
period.
Avg.
Avg.Relativity
Relativity
' =
' =
Base
Base
Rate
Rate
' '
Limited
Limited Average
Average Severity
Severity
𝐿𝐿'𝐿𝐿' ∑2'*-
∑2'*-
min(𝑥𝑥
min(𝑥𝑥 ' , 𝑢𝑢)
' , 𝑢𝑢)
Adj.
Adj.𝐿𝐿z'𝐿𝐿z=
' = 𝐿𝐿𝐿𝐿𝐿𝐿(𝑢𝑢)
𝐿𝐿𝐿𝐿𝐿𝐿(𝑢𝑢)==
Avg.
Avg.
Relativity
Relativity
' ⋅'Exposure
⋅ Exposure
' ' 𝑛𝑛 𝑛𝑛
Ç∑Ç∑
B CD
B CD𝑥𝑥'𝑥𝑥É '+
É+ 𝑢𝑢 𝑢𝑢
⋅ 𝑛𝑛⋅D𝑛𝑛D
Adj.
Adj. 𝐿𝐿z'𝐿𝐿z' 𝐿𝐿𝐿𝐿𝐿𝐿(𝑢𝑢)== ! !
𝐿𝐿𝐿𝐿𝐿𝐿(𝑢𝑢)
Indicated
Indicated
Relativity
Relativity
' =
' =
𝑛𝑛 𝑛𝑛
Adj. 𝐿𝐿z>?@A
Adj. 𝐿𝐿z>?@A where
where 𝑛𝑛D𝑛𝑛is
D is
the
the number
number ofof losses
losses
greater
greater
than
than𝑢𝑢.𝑢𝑢.
Indicated
Indicated Avg.
Avg. Rate
Rate Note:
Note: 𝐿𝐿𝐿𝐿𝐿𝐿(𝑢𝑢) is is
𝐿𝐿𝐿𝐿𝐿𝐿(𝑢𝑢) thethe empirical
empirical version ofof
version E[𝑋𝑋 ∧∧
E[𝑋𝑋 𝑢𝑢]𝑢𝑢]
Indicated
Indicated
Base
Base
Rate
Rate
==
Indicated Avg.
Indicated Relativity
Avg. Relativity Increased
Increased Limit
Limit Factor
Factor
Credibility-Weighted
Credibility-Weighted Relativities
Relativities 𝐿𝐿𝐿𝐿𝐿𝐿(𝑢𝑢) ++
𝐿𝐿𝐿𝐿𝐿𝐿(𝑢𝑢) 𝑅𝑅𝐿𝐿𝑅𝑅𝐿𝐿
D D
𝐼𝐼𝐼𝐼𝐼𝐼
𝐼𝐼𝐼𝐼𝐼𝐼==
New
NewRelativity
Relativity==
𝑍𝑍(Indicated
𝑍𝑍(IndicatedRelativity)
Relativity) 𝐿𝐿𝐿𝐿𝐿𝐿(𝑏𝑏)
𝐿𝐿𝐿𝐿𝐿𝐿(𝑏𝑏)
++ 𝑅𝑅𝐿𝐿𝑅𝑅𝐿𝐿
E E

++ (1(1
−−𝑍𝑍)(Current
𝑍𝑍)(CurrentRelativity)
Relativity) • •𝑏𝑏:𝑏𝑏: original
original limit
limit
• •𝑢𝑢:𝑢𝑢: increased
increased limit
limit
Note:
Note: Risk
Risk load
load (RL)
(RL) function
function will
will bebe provided
provided byby the
the question
question if if
needed.
needed.
Indicated
Indicated Relativity
Relativity == 𝐼𝐼𝐼𝐼𝐼𝐼
𝐼𝐼𝐼𝐼𝐼𝐼

Loss
Loss Elimination
Elimination Ratio
Ratio
𝐿𝐿𝐿𝐿𝐿𝐿(𝑑𝑑)
𝐿𝐿𝐿𝐿𝐿𝐿(𝑑𝑑) −− 𝐿𝐿𝐿𝐿𝐿𝐿(𝑏𝑏)
𝐿𝐿𝐿𝐿𝐿𝐿(𝑏𝑏)
𝐿𝐿𝐿𝐿𝑅𝑅
𝐿𝐿𝐿𝐿𝑅𝑅
F F==
𝑥𝑥̅ 𝑥𝑥̅−− 𝐿𝐿𝐿𝐿𝐿𝐿(𝑏𝑏)
𝐿𝐿𝐿𝐿𝐿𝐿(𝑏𝑏)
• •𝑏𝑏:𝑏𝑏: original
original deductible
deductible
• •𝑑𝑑:𝑑𝑑: increased
increased deductible
deductible
Indicated
Indicated Relativity
Relativity == 1−
1− 𝐿𝐿𝐿𝐿𝐿𝐿
𝐿𝐿𝐿𝐿𝐿𝐿

Reinsurance
Reinsurance
• •Facultative:
Facultative:Used
Used forfor
ceding
ceding individual
individualrisks.
risks.
• •Treaty:
Treaty:Used
Usedforfor
ceding
ceding allall
risks
risks
inin
a specific
a specificline
line
oror
class
class
ofof
business.
business.
• •Quota
Quotashare:
share:Both
Both parties
partiesshare
sharea percentage
a percentage ofof
the total
the risk.
total risk.
• •Surplus
Surplus
share:
share:Both
Both
parties
partiesshare
share a percentage
a percentageofof
the
the
total
total
risk
risk
above
above
the
theretention
retentionlimit.
limit.
• •Excess
Excess
ofof
loss:
loss:
The
Thereinsurer
reinsureris is
responsible
responsible
forfor
the
the
claim
claim
amounts
amounts
exceeding
exceeding the
the
retention
retention
limit.
limit.

Copyright © 2021 Coaching Actuaries. All Rights Reserved. 8

www.coachingactuaries.com Personal copies permitted. Resale or distribution is prohibited.