ILA-LPM B Flashcards

Question

Which risk classes have separate mortality studies?

Answer 1

1. non-routine UW 2. conversions ⇒ usually higher mortality 3. sub-standards 4. non-forfeiture (ETI, RPU) 5. multiple-life

Answer 2

A + N = W + D + B ## Footnote of lives... A = beginning of year N = enter during year W = lapse during year D = die during year B = alive at end of year

Answer 3

* Policy level - sum up individual exposures * Aggregrate - assume new issues and WDs occur mid-year (s=r=0.5)

Answer 4

* multi-year studies - better credibility * 5-yrs good; too many years hide trend * lower for amount-based studies * A/E ratios - credibility & tracking trends

Answer 5

Total mortality = weighted average of: 1. mortality of **"select lives" that lapse** in duration r 2. mortality of lives that **stay w/ original policy** **Key pricing concept:** If you don't account for selective lapsation, you will under-price the insurance product

Answer 6

* healthy lives more likely to lapse than unhealthy - especially w/ high renewal premiums * mortality of those left begins to increase by "the conservation of deaths" principle

Answer 7

1. product design 2. distribution channel 3. policy size 4. premium mode 5. product type 6. conservation program effectiveness

Answer 8

* term: high (shock) lapses when ART period begins * annuities: high (shock) lapses after SC period * permanent insurance: high 1st year lapses

Answer 9

* lapses generally occur on premium due dates * flexible premium: assume uniform monthly

Answer 10

* small policies: high early lapses * large policies: high later lapses

Answer 11

* brokerage: higher lapses * agency: higher lapses if agent quality is poor

Answer 12

* Deferred annuities: more sensitive to lapse rates than life insurance

Answer 13

* same as mort study - death and lapses trade places * lapses get full year of exposure w(x) = W / (A + (1 - r)N - (1 - s)D)

Answer 14

* failure to pay premium (term) * full/ partial cash surrender * policy loan \> CSV * nonforfeiture transfers - ETI, RPU * term conversions * deviations in actual prem as a % of target prem (UL)

Answer 15

Investment Assets: * deterministic * portfolio average * investment generation * stochastic * useful for more important risk * rates vary w/ time, asset class, quality, & credit risk

Answer 16

* modeled as assets or negative liabilities * net of policy loan expenses * utilization rate

Answer 17

I = Ai + (B - A - I)(i/2) i = 2I / (A + B - I) A = BoY B = EoY C = net CFs (assume mid-year)

Answer 18

r = (B - A - C) / (A + C/2) If done daily, set C = 0 A = BoY B = EoY C = net CFs (assume mid-year)

Answer 19

* Direct (vary w/ sales) * commissions * premium taxes * UW * Indirect (express as per policy, % of premium, or per unit) * OH * Both * maintenance * acquisition * entering new LOB

Answer 20

1. **Goal**: develop a policy count base for per policy expenses 2. **How**: Count the number of policy years that **start in the study's calendar year** * BOYs "crossed" in CY = A + N -W/2 * Mid-years "crossed" in CY = A + N/2 - W/2 * Exposure count * (A + B + N)/2 for **beginning of year expenses** * (A + B)/2 for **mid-year expenses** * assumes WDs & new issues occur mid-year

Answer 21

1. transaction count (# of premium payments) 2. transfer costs (employee benefit cost per employee) 3. employee time spent 4. index-based allocation (policy count or premium)

Answer 22

combinations of data dimensioned by issue age, sex, smoker/nonsmoker, policy year, etc.

Answer 23

* decrement = probabilities from 0 to 1 * mortality, morbidity, lapse, etc. * utilization = NOT probabilities (can exceed 1) * WD, option election rates, etc.

Answer 24

* *l_x* terms become *l_x* x avg DB **in force** * *d_x* terms become *d_x* z avg DB **paid on death** * *w_x* terms become *w_x* x avg DB on withdrawn policies * sum amount exposures and calculate *q_x*'s in the usual way

Answer 25

* multiply each life year exposure by the policy's DB * sum amount exposures and calculate q_x's in usual way

Answer 26

Amount-weighted *q* will be \> life-weighted *q* if average DB paid on death \> average DB in force

Answer 27

* compare actual mortality/lapse to expected * develop best estimate assumptions as a multple of expected amounts * Best estimate rate for age x = (A/E) x q_x^e * valuation, risk management, financial planning, etc.

Answer 28

* f_x = n_x/E_x = avg claim frequency* * s_x = C_x/n_x = avg claim amount* * *n_x* = # of claims incurred at age *x* * *C_x* = total claim amount incurred at age *x* * *E_x* = central exposure (or initial)

Answer 29

* u_x = s_x/M_x* * M_x* = max w/d allowed * s_x*= average size (severity) of w/d

Answer 30

* s_x = W_x/n_x* * W_x* = total actual w/d amount * n_x* = # of contracts who took a w/d

Answer 31

* f_x = n_x/E_x* * E_x* = count-based exposure (deaths and lapses) * n_x* = # of contracts who took a w/d

Answer 32

* E_x = l_x - d_x - w_x* * w_x*= # of lapses

Answer 33

* **large amounts can distort numerator** * not easy to fix (cap max amount allowed in study) * **risks/behaviors can differ significantly by amount** * small policies have less commitment * large policies used to defraud * solution: add band size to cell

Answer 34

* higher credibility * less distortion cause by reporting lags * populate many more cells * allow study of trends for study-year/CY

Answer 35

* **death claims** - not reported for months * **LTCI claims** not reported til after EP * **"shoe-boxing"** - lag between direct insurers and reinsurers * **life annuities** sometimes continue paying after death

Answer 36

* wait a few months after study period before gathering data * IBNR claim estimates from previous studies

Answer 37

1. lapses often vary by month 2. prorated mortality rates tend to underestimate deaths early in year at old ages 3. LTCI claims immediately following claim

Answer 38

1. shorter interval 2. adjustment factors 3. constant force mortality for old ages

Answer 39

* **_ok_ when _evenly distributed_ over year** * common for biologically-driven rates (mortality, morbidity, etc.) * can allocate PYs between CYs * **Otherwise, partial policy years will distort results** * behavioral rates not usually evently distributed * lapse, withdrawal, option election, etc. * do not allocate PYs between CYs

Answer 40

* assumes decrement is **_uniformly_ distributed** * **allocates exposure/decrements** by both PY & CY * produces **same total PY exposure as annual method**, but CY allocation differs * **exposure missed by intial partial year** (so can extend past study period)

Answer 41

* consistent with **Balducci** * exposes decrements to **end of next anniversary** (even beyond study period) * **disadvantage**: overstates rates in first partial year; understates rates in final partial * **expsoure extended** beyond study period: * b/c rate is probability * exposure missed by intial partial year

Answer 42

1. calculate annual exposure 2. 1st half exposure = PY *t* exposure in CY * deaths/surviving policies from anniversary to end of CY * other decrements from anniversary to decrement date 3. 2nd half exposure = total - 1st half exposure * allocated to next CY * In plain English: switch the partial year formulas in the annual exposure method* * *Move final partial year deaths to next study period’s first partial year*

Answer 43

* deaths in premium grace period (30-60 days) * compromised and denied claims * study cariables changing over time * reinsured amounts * substandard and uninsurable lives

Answer 44

* actual DB paid is net of overdue premiums, but use full DB in mortality study * expose lapses to **beginning** of grace period

Answer 45

* admin system changes * may change plan codes or billing frequencies (distorts exposure)

Answer 46

* don’t count as death claims in mortality study

Answer 47

* can study amounts net of reinsurance if material enough

Answer 48

* typically excluded

Answer 49

1. multiple payments that can last months or years 2. usually paid monthly 3. EP before claim payments start (up to 1 year) 4. usually limited in some way (e.g. max age for DI, max total benefits, etc.) 5. paid when insured meets certain conditions 6. claims stop when insured recovers, but can start again

Answer 50

(LTCI only) Benefit Utilization Rate = Actual Benefits / Max Benefits

Answer 51

* separate studies for recovery and death * DI claims terminate when proof of diability not submitted * LTC claims terminate when requests for reimbursement stops * termination rates are highest and most volatile for new claims

Answer 52

Average length of claim = total months of claims/# of claims Average cost per claim = total claims paid/# of claims

Answer 53

* claim incidence rates = annual probabilities * usually done on "all claims" * usually based on annual exposure method

Answer 54

1. **EPs** - claims are usually reported after EP 2. **partial DI benefits** - reduces avg cost per claim 3. **recovery followed by relapse** within 6 months 4. **claim settlements** - lump sum paid in lieu of future monthly

Answer 55

1. EP - 30, 60, 90, 180, and 365 2. some benefits indexed to inflation 3. mortality rates do not follow any established table 4. diagnosis - must capture (drives claim length) 5. claim data not always organized - requires work

Answer 56

* **deferred annuity utilization rates** * GMxBs: compare benefit paid to the amount payable in absence of GMxB * **contract year data challenges** * data usually monthly * common approach: convert account balances to contract year basis using a 13-month average

Answer 57

* from injury lawsuits * one of more lump sums (not lifetime payouts) * mortality higher than immediate annuities * amount exposure - base on reserves or premium paid

Answer 58

* self-selection - people buy IAs when they think they are healthier than average * choice of immediate annuity may indicate health * larger IAs tend to have lower mortality * amount exposure should be based on monthly payment, not reserves * mortality improvement is a key consideration

Answer 59

1. develop table 2. identify table dimensions (exploration techniques) 3. populate table (graduation/interpolation/modeling) 4. extend and project rates 5. review and adjust rates 6. finalize

Answer 60

1. review previous studies 2. clarify puprose of table 3. ensure confidentiality of each contributor's data 4. review available data (homogeneity, credibility, etc.)

Answer 61

* **data call**: formal request for data from **contributors** * **simpler** - more successful * document what is needed for each **data item** (columns) * define data **structure** (e.g. relational database) * create **detail records** (rows) * **document** all table relationships

Answer 62

1. incomplete data 2. terminology may vary by contributor 3. data may not arrive on time 4. wrong or improperly formatted data 5. time lags in reporting 6. lack of resources to transform data

Answer 63

1. **study anniversary date -** bday, policy anniversary, etc. 2. **age basis** (ANB, ALB, etc) 3. **experience study summary records** 4. **count-based and amount-based rates** (*q_x*'s, etc.) 5. **expected results** (A/E ratios) 6. **summary results for each record** (event counts and amounts)

Answer 64

1. assume all ages based on most common method 2. recalculate on common basis for each record 3. weighted-average age

Answer 65

1. **acquire** data 2. data **validation**, preliminary **exploration**, **outlier** analysis 3. data **visualization** and preparation 4. **analytical approach**: exploratory vs. advance analytics 5. **model creation** and assessment * select model(s) w/ explanatory ability, predictive power, and implementation ease 6. **select final model** 7. minimize table **dimensions** 8. replacing grids w/ **factors** if possible THINK EXAM PA!

Answer 66

The modeling process: 1. model fitting: select variables w/ lowest p-value 2. avoid confounding 3. create transformed variables, functions, interaction terms, or stratification AGAIN - THINK PA!

Answer 67

* **Graduation** is a mathematical process that smooths an array of rates * assumes "true" rates follow continuous curve * _required_ when rates _don't_ come from model * can still be used on modeled rates * **goal: balance fit and smoothness**

Answer 68

* produces same total events as ungraduated * parameters control amount of smoothness * input fit, table fit, smoothness

Answer 69

* **non-credible** ("incredible") rates distort graduated rates * smoothness calc **weights** credible and non-credible rates equally * biggest problem: very young and very old ages * increasing **credibility comes at cost of graduation** * grouping rates by age (e.g. quinquennial) * grouping data by calendar or policy years * **possible solutions**: use exposure-weighted averages for graduation

Answer 70

1. **Collect and populate graduation input** 2. **Review and adjust input as needed** 3. **Run graduation process** for 3 parameters: _input fit_, _table fit_, _smoothness_ * Set parameter value and run the graduation algorithm * Review/graph/evaluate results * Repeat until satisfied with parameter 4. **Run a final graduation** using the best fit and smoothness parameters

Answer 71

1. overall average size 2. average size for lives contributing to *q_r* 3. allocate average size between 2 most common sizes in size groups (can be used to develop factors for 1 & 2)

Answer 72

* **1-dimensional methods** * linear * cubic spline * log-linear or log-cubic (good for mort) * **2-dimensional methods** * bilinear * bicubic (smoothest)

Answer 73

Often necessary for very young or old ages where credibility is low 1. **use rates from existing _credible_ table** * use slope as guid * grade from study's rates to existing table's rates 2. **other data sources** * SSA for old ages 3. **formulas** * should reproduce rates for nearby credible ages

Answer 74

1. nonsensical (mort rate \> 1) 2. data had poor credibility 3. nonsensical pattern 4. suspicious differences in arrays of rates 5. very wide CIs

Answer 75

1. define **relationships to be enforced** 2. define when **relationships will be checked** 3. create spreadsheets to **check relationships** (automation) 4. check **S&U mort rates** 5. adjust rates to **enforce relationships**

Answer 76

1. **historical data** - consistent population 2. **mortality trends** - difficult to estimate 3. connect **cause and effect** 4. different types of **projections**

Answer 77

1. mid-point of experience data to final table effective date 2. beyond effective date at single rate of improvement 3. beyond effective date at a varying rate of improvement

Answer 78

* **reserves and nonforfeiture values** * prescribed industry tables affect prescribed stat reserves and nonforfeiture values * PBR can be impacted by internally developed rates * impact can vary a lot from company to company * **premiums and PH dividends** * could be impacted by new industry or internally-developed tables

Answer 79

1. w/ or w/ out projected trend factors 2. w/ or w/ out valuation loading factors 3. multiple age definitions (ANB, ALB, etc.) 4. unisex 5. relative risk versions for UW purposes

Answer 80

1. develop **experience table** - soley on experience data 2. develop **valuation basic table (VBT)** 3. apply loading factors to experience table to create a **loaded experience table** 4. create **CVT** (used for stat reserves)

Answer 81

* for both LFM & Buhlmann, credibility estimate: Estimated Company A/E = Z x (Company A/E) + (1 - Z) x (Overall A/E) * Z = credibility factor ranging from 0 to 1 * Z = 1 means company experience fully credible * Calc for Z different under LFM and Buhlmann * overall A/E ratio based on average of all companies in study * lower Z is, more weight placed on overall experience

Answer 82

* based on CI and only uses data from particular company being studied to determine the credibility factor

Answer 83

* based on empirical application of Bayesian stats * starts w/ initial or prior distribution based on past data, professional experience, or option * observed results used to formulate predictive or posterior distribution * considers variance between insurers, not just insurer specific like LFM * Z calc more complex

Answer 84

* **Buhlmann** considers 2 sources of variance, LFM only 1st: 1. company-specific 2. among companies * **LFM only** needs data from company studied (simpler) * **Both**: * calculate estimated company A/E * assume constant overall A/E ratio * additional adjustments maybe needed based on actuarial judgment * Comparison for 10 insurers surveyed: * **count-based** A/E estimates **similar** * **amount-based** A/E estimates **differed** more

Answer 85

1. Commonly accepted/well established 2. easy to produce results 3. management familiar with ⇒ can apply to management decisions

Answer 86

1. project **scope** 2. **data** collection and validation (MOST IMPORTANT) 3. **initial factor** anlysis 4. model **building** 5. model **validation** 6. final **calibration** 7. **implementation**

Answer 87

* clean data and consider variables that may not be obvious

Answer 88

* multiplicative or additive * statistical tests, business knowledge, and interactions to include

Answer 89

* compare A/E ratios * use hold-out samples

Answer 90

* do univariate analysis * then remove highly correlated variables

Answer 91

* refit model using all data and make final adjustments

Answer 92

* use model results to set assumptions, do underwriting, etc.

Answer 93

* (target variable, LOB, timeline, resources) * Assemble team of technical and business experts

Answer 94

Traditional techniques don’t use these but **predictive analytics** could. . . * **UW type** * could use process results (BP, LDL, BMI) * limitation: data availability * **geo-demographic variables** * categorize socio-economic composition of a particular zip code * apply multipliers that vary by face amount, duration, or distribution channel

Answer 95

1. **Interactions** of factors better understood 2. better **credibility** for specific factor combinations 3. **standardizes** all factors ⇒ isolates "true effect" of each 4. backed up by **statistical tests**

Answer 96

* assess NB value at more **granular** level * allows management to make more **specific changes** if needed * revise benefits, increase fees, change producer compensation

Answer 97

* significant expertise required * risks: misinterpret data and flawed models

Answer 98

* lack of data at old ages (85+) * possible solutions: create trend lines or adjust standard models * not doing experience studies at all * some companies lack resources * pricing models may not support the more refined assumptions

Answer 99

1. collect and organize data 2. prepare data 3. build model 4. final adjustments

Answer 100

1. relationship with target variable 2. statistically signifcant 3. correlation vs. cost to obtain 4. regulatory/legal concerns 5. 12-18 months of data

Answer 101

* **Traditional** * app data * medical information bureau (MIB) * motor vehicle record (MVR) * electronic Rx profile * traditional medical UW results * **Non-traditional 3rd party data** * credit scores (FCRA) * marketing data

Answer 102

1. generate variables 2. exploratory data analysis 3. variable transformation 4. partition into 3 sets

Answer 103

* synthetic vs. disease-state

Answer 104

* distributional vs. univariate analysis

Answer 105

1. Group values into buckets 2. Replace missing values 3. Reduce effect of extreme values/outliers 4. Convert variables into numerical values to capture trends

Answer 106

1. Train 2. Validation 3. Test

Answer 107

* **layering on existing UW guidelines** * example: exceptions for rare conditions * **decision trees** * used to narrow applicant pool into good/bad risks

Answer 108

1. PH can't manipulate 3rd party data 2. model relies on data in total, not individual 3. allow reduced UW requirements for some 4. randomly traditionally UW some applicants (for checks)

Answer 109

1. collection of consumer data 2. marketing data not subject to FCRA 3. social views on privacy 4. compliance and legal staff should review variables

Answer 110

* main idea: *q_x* = *f*(credit behavior) * Credit behavior indicates an individual’s conscientiousness * conscientious people tend to have better credit history ⇒ lower mortality

Answer 111

* Engage in healthier behavior/environments * Healthier friendships * Better education, career, and higher incomes * Have less unhealthy stress

Answer 112

* lowest conscientious US adult mortality = 3.2x highest conscientious * falls to 1.6x after adjusting for smoking, alcohol, and waist size

Answer 113

* **adjacent markets** utilize similar data/approaches * heavily **regulated** - allows for consumer dispute resolution * updated in near **real time** * used to develop model that is fit for purpose: **mortality** * **credible population** - allows for a separate holdout sample * **available** for vast majority of adult population

Answer 114

The score is a function of 25 credit attributes across 4 categories: 1. **credit-seeking activity** - recency/frequency of credit inquiries/trades 2. **credit tenure** - credit history length, active trade count, months since oldest trade 3. **severity/frequency of derogatory credit info** - bankruptcies, collections 4. **credit usage** - utilization percentages, usage patterns, recency

Answer 115

* 44M: **training** data used to create model w/ multivariate logistic regression * 30M: **test/validation** data to prevent overfitting * 18M: **holdout** data for mortality validation

Answer 116

* 1998 TransUnion credit archive: 92M useable records * originally 175M (roughly 90% US pop) * useable set only includes adulst w/ a SSN, credit history, and age 21-70 * 2011 SS Death Master File (death data)

Answer 117

* low correlation * high predictive power * stable over time * non-gameable

Answer 118

* TrueRisk Life score for each individual = 1 to 100 * 1 = best credit-based score possible (100 is worst)

Answer 119

* **credit history generally improves w/ age** * A/Es still increase within each TrueRisk Life score slightly w/ age * ages 60+: health factors play a bigger role

Answer 120

* **A/E ratios increase w/ TrueRisk score (worsening credit)** * worst 5% of scores mortalty more than 6x the best 5% * A/E ratio curve smooth and monotonically increasing ## Footnote **Credit segmentation does not wear off over time**

Answer 121

* Compared to full & non-medical UW * **FU business** skews toward l**ower** TrueRisk Life score (better credit) * **non-medical** skews toward **higher** * A/E ratios **increase** with score for both UW types * early duration **lapse** rates **increase** w/ score

Answer 122

**TrueRisk Life is subject to the FCRA** * Governs collection, assembly, and use of consumer report information * Model does not use: * Credit card transactions * Social media * Income * Race * Criminal records or court filings * Property records * Religion or national origin * IRS data * Education * Checking/savings account information

Answer 123

* **score ranges from 200 to 997 (worst to best)** * Population studied: 8 million insurance shoppers ages 18–90 * Study period: 2006–2010 * based on public records, credit information, motor vehicle history * Death sources: SS Death Master File, state records, other * Expected mortality basis: 2008 VBT

Answer 124

* **Risk Classifier stratifies mortality risk across age, gender, and duration** * Mortality risk decreases with score within age bands and gender * Older ages (70+) have less differentiation by score (health plays a bigger role) * Credit tends to improve with age * Differentiation does not wear off with time

Answer 125

* **Better public records and credit indicate better mortality** * Mortality risk increases as past due balances increase * Other attributes that result in higher mortality: * Derogatory public records (bankruptcies, liens, judgments) * Felony and criminal convictions * People with professional licenses have lower mortality

Answer 126

* **Risk Classifier scores generally increase with wealth**

Answer 127

* **Mortality risk _decreases_ as the Risk Classifier scores increase** * People with better credit/behavior have lower mortality * Exposure is concentrated in the middle range of the scores * Exposure is very low at the best scores (900+)

Answer 128

1. Forced Method 2. Blended Method 3. Pattern Method 4. Less-Than-One Method

Answer 129

* Choose an ultimate age * No changes to other qx ’s * At ultimate age, qx may be 1.000

Answer 130

* qx ’s continue normally until they reach 1.000 * Age where qx = 1.000 is ultimate age

Answer 131

* Choose ultimate age * Grade qx ’s to 1.000 at ultimate age * i.e. “blend” the qx ’s to 1.000

Answer 132

* Choose ultimate age * Set ultimate qx = 1.000 at ultimate age * No changes to other qx ’s

Answer 133

* Difficulties with ultimate age and mortality levels * Little data at old ages * Questionable accuracy

Answer 134

* Female slope higher at old ages (catching up to male) * Mortality increases at a decelerating rate from age 100 to 110

Answer 135

1. Increased product **flexibility** 2. More **investment** components 3. **Volatility** in financial markets 4. Sophisticated financial **reporting and regulation** 5. Social and technological **advances** 6. Emerging **behavioral economics** research

Answer 136

* **Policyholder behavior risk: Non-diversifiable and non-hedgable** * Elective, _unlike_ mortality risk

Answer 137

* **Product design** – early mistakes were made: * VA GMDB dollar-for-dollar withdrawals, LTCI lapse rates set too high * **Reserves & capital** – Premium paid, investment elections, fund transfers, etc. * **ALM** – Liability durations change as policyholders react to changing interest rates * **ERM** – Tail risk and other operational risk

Answer 138

1. Lack of time 2. Information asymmetry 3. Curse of knowledge 4. Data challenges 5. Human decision making is not fully understood (but improving)

Answer 139

* (but improving) * Behavioral economics, Big Data, computing power, dynamic modeling

Answer 140

* Lack of data, credibility, and causal linkages * Systems weren’t designed with policyholder behavior in mind * Data not in required format

Answer 141

* Significant gap between policyholder and actuarial product views

Answer 142

* Insurers have incomplete information about policyholder * Policyholders have incomplete understanding of products

Answer 143

Time is spent preparing data, not analyzing it

Answer 144

* **Own experience with the highest credibility:** * Surrenders * Lapses/premium persistency * Loan utilization * Dividend elections * **Own experience with the lowest credibility:** * Mortality shifts due to lapses deviations * Shock mortality and lapse * Index elections/transfers

Answer 145

* **Dynamic Assumptions** – vary with external factors in model * common uses: ALM, risk management

Answer 146

* Surrenders (UL, SGUL, annuities) * SGUL driver: in-the-moneyness * UL and annuity driver: credited rates * VA and EIA living benefit utilization * VA withdrawals and transfers * Others: loan utilization, funding, partial withdrawals, anti-selection

Answer 147

* **Static Assumptions** - does NOT vary with external factors * Common uses: pricing, valuation, forecasting

Answer 148

how optimally policyholders maximize their contract values

Answer 149

* most companies assume less than 100% (consistent with history) * most companies assume “neutral” to “very efficient” * may increase with sophistication of products/policyholders * Half of companies vary with product * Some companies vary with distribution channel

Answer 150

**Formal governance process** – agreed-upon / well-communicated procedure for. . . 1. **Updating** experience data 2. **Reviewing** changes in experience 3. **Recommending** modifications of assumptions 4. **Discussing** consistency of assumptions across products/areas 5. **Sign-off** on recommended changes 6. **Validating**, updating, and reviewing actuarial models

Answer 151

* **Common techniques**: A/E analysis, dynamic validation * **Newer techniques**: back-testing, 3rd party review, stress/scenario testing

Answer 152

Data availability and credibility

Answer 153

1. Traditional actuarial modeling (by far most popular) 2. Predictive modeling 3. Monte Carlo simulation

Answer 154

* Balance dynamic assumption complexity with value added * Identify (in)consistencies across lines/areas

Answer 155

* Review data sources * Consider: * Customer service and web-based data * Applicability of in-force block experience to new products * Policyholder behavior under extreme scenarios * Apply predictive modeling / statistical techniques * Centralize internal data sources

Answer 156

* **Implement**: * Formal process for experience data updates * Review procedure * **Establish cross-functional**: * Discussions about inconsistencies * Data analytics repository * **Experiment with advanced statistical techniques**

Answer 157

* Establish formal process * Document changes * “Model Sherriff”

Answer 158

1. Identify common shortcuts used in consumer decision-making 2. Explain external variables affecting consumer decision-making 3. Explain market inefficiencies 4. Offer insights into product design and marketing

Answer 159

1. Relative choices 2. Reliance on defaults 3. Mental accounting 4. Framing 5. Status quo bias

Answer 160

people tend to stick with prior choices

Answer 161

how choices are labeled can influence consumer decisions

Answer 162

tendency to create artificial budgets covering different categories of spending and saving

Answer 163

tendency to rely on assumptions set in the past

Answer 164

tendency to make (sometimes irrational) decisions by comparing available choices

Answer 165

1. Love of free 2. Anchoring 3. Endowment effect 4. Hyperbolic discounting

Answer 166

people tend to heavily discount the importance of future events (procrastinate

Answer 167

people don’t like to give up something they already have

Answer 168

tendency for people’s numerical estimates to be influenced by suggested numbers, even if completely irrelevant

Answer 169

people have a disproportionate love of free things

Answer 170

1. Risk aversion 2. Overconfidence 3. Loss aversion 4. Self-control facilitation 5. Hot and cold states 6. Self-herding

Answer 171

future decisions may be made to justify past behaviors

Answer 172

being in a good or bad mood can lead to irrationally optimistic or pessimistic decisions

Answer 173

willingness to accept less freedom in order to facilitate saving

Answer 174

bias to hold on to “losers” too long to avoid realizing losses

Answer 175

overestimating one’s ability to make correct choices

Answer 176

primary reason policyholders buy insurance: people don’t like risk

Answer 177

people do what they see other people do

Answer 178

1. Decision Shortcuts 2. Value Assessment 3. Emotional Impacts 4. Social Impacts

Answer 179

* “Masculine” societies buy more life insurance than “feminine” societies

Answer 180

* **Loss-aversion** – buying life insurance feels like a “loss” (buyer won’t get benefits) * **Overconfidence bias** – people underestimate risk of death (especially men!) * **Hot and cold states** – people are more likely to buy insurance after a tragedy

Answer 181

* Solutions for insurers: * Stretch the **time horizon** over which risk is measured * **Frame** life insurance as an investment in loved ones * Associate **positive emotions** with life insurance (e.g. par WL becomes “free”)

Answer 182

* life insurance is “sold, not bought” * people **anchor** on unlikely-to-happen current health risks * Insurers show/quantify the **value** over a lifetime

Answer 183

* Life insurance is associated with a negative event (death) * **Framing:** agents should sell a policy as a “disciplined, mandatory savings plan”

Answer 184

* annuitization involves complex decisions * **Mental accounting** – consumers that view annuitization as a gamble will segregate the annuity * **Framing** – people annuitize more often when they see it as consumption rather than investment * **Default decision** – many people opt out for a lump sum * Possible solution: let people “test drive” the annuity

Answer 185

* people **hyperbolically discount** future annuity payments (devaluing them) * **endowment effect** – people don’t want to give up their assets * **availability heuristic + conjunction fallacy** – people underestimate longevity risk

Answer 186

* need to minimize the association between profit and death

Answer 187

* **Loss aversion** – people are unwilling to “gamble” on “losing” their savings * People don’t like the word “annuity”

Answer 188

* **Interest rate hypothesis** * High guaranteed credited rates lead to lower lapses * Lapses increase with market interest rates and stock returns * **Emergency funds hypothesis** * Policyholders surrender their policies due to financial distress **These are NOT based on behavioral economics**

Answer 189

* Premiums: expenses or savings? * CSV: spending money or asset accumulation? * Lapses **decrease** as savings mindset **increases** * Marketing and product features can influence policyholder mindset

Answer 190

* Policyholder’s “fair value” of current contract may be irrational * May lapse if other contracts’ features seem more attractive * Insurer should: * Be careful in how product is presented * Be aware of other products policyholders will compare against

Answer 191

* **“Value” is a relative concept** * Influenced by emotions rather than rational decisions * Tone of questions can influence answers * “Live to X years” vs. “die by X years” * **Hyperbolic discounting** * Policyholders are likely to postpone financial planning * **Love of free** * Policyholders may prefer self-supporting products that seem “free”

Answer 192

* **Life:** Pay premiums but never see death benefit * **Annuity:** Avoid relying on family members in retirement

Answer 193

* Anxiety may influence lapse and withdrawal decisions * Marketing and other communications could trigger hot states

Answer 194

* Some European countries require higher CSVs * Increasing policyholder options may not be in policyholders’ best interests

Answer 195

* Policyholders may hold a “loser” VA contract (but guarantee may be a “winner” if ITM) * LTCI: policyholders persist to avoid pain of “losing” premiums paid

Answer 196

Lapses **increase** when risks go away (pay off mortgage, child goes to college)

Answer 197

* **Decisions are affected by family, friends, and other social contacts** * **Bandwagon effect** * May affect mass lapse risk and liquidity strain

Answer 198

* Policyholders. . . * May follow target premium illustrations * Tend to pay same premium as last period * Many actuaries assume recent patterns experience will continue

Answer 199

* Increases policyholder inertia and procrastination

Answer 200

* Numbers in marketing materials may influence funding * Avoid unintended anchoring in communications

Answer 201

* Decisions to pay premiums may trigger “pain” * Higher premium payment frequencies = more pain per year * Solution: frame premiums as savings rather than expenses

Answer 202

* Policyholders may continue paying a premium because it’s what they did in the past

Answer 203

* A savings perception may delay taking withdrawals * Tendency to see annuities as a “first source” of retirement income

Answer 204

* Communications may signal when they should begin withdrawing * Product design may send signals (e.g. setting “full-benefit age,” bonuses)

Answer 205

* Tendency to **underestimate retirement date** * Tendency to base financial planning on **“best case” scenarios** * Actual product use may be **very different** than planned * Tendency to **underestimate future expenses**

Answer 206

* Policyholders may not realize how **fungible** money is * Withdrawal patterns may be tied to **“mental buckets”** * Understanding how policyholders perform mental accounting may **reduce uncertainty**

Answer 207

* Withdrawals may be **based on amounts endorsed by experts** or social norms

Answer 208

* Withdrawals may be influenced by **irrelevant numbers** shown to policyholders

Answer 209

* “1/n rule” and “conditional 1/n rule” * Tendency to allocate money equally over all or a subset of choices * Overwhelmed policyholders seek safer options * Number and presentation of investment options has a large impact on allocations and policyholder satisfaction

Answer 210

* “Retirement account” vs. “asset accumulation tool” * More form lines ⇒ more selections made by policyholders

Answer 211

* Policyholders may allocate “old money” vs. new money

Answer 212

* Tendency to follow insurer’s suggested allocations * Relying on defaults reduces policyholder’s future regret

Answer 213

* Tendency to put off changing allocations for long periods of time * Tendency to leave designated “savings” alone despite better opportunities * Insurer considerations: balance profit opportunities with regulatory scrutiny

Answer 214

* Risk averse policyholders more likely to choose bond funds over stock funds

Answer 215

* Tendency to hold “loser” stocks to avoid realizing losses * Policyholder’s decision environment is different that stock investor’s: * VA reallocations don’t trigger taxes * Insurance guarantees may offset account losses * Insurer should: * Present distribution of outcomes over a long-term horizon * Reduce feedback frequency

Answer 216

* May result in high levels of trading (men suck!) * “Law of small numbers” – drawing inference from small data sets * “Availability bias” – overweighting salient, personal, or friends’ experience

Answer 217

* Market signals trigger emotional states (fear, anxiety, optimism) * Insurers/advisors may also stimulate policyholder states

Answer 218

* **Bandwagon effect** * Your beliefs ≈ your friends’ beliefs * More pronounced during unusual changes in society, technology, or economy * May result in poor market timing

Answer 219

* Annuitization usually requires an active decision (opt-in) * Insurers could make annuitization a default option

Answer 220

* Annuitization utilization may **increase** with: * Marketing and education for the need for retirement income * Increased focus on income stream, not account balance * Avoid making the income stream seem like a “loss” * Offer trial periods

Answer 221

Certain income streams create mental ease

Answer 222

Income stream more likely to seem like a “loss” if policyholder thinks it lasts only a few years

Answer 223

* **Aggregate level modeling** – does not consider socio-demographic, attitudinal, or behavioral factors * Causal factors behind lapse decisions are lost * Removing variation in data hurts model * **Rational approach** – does not account for social, cognitive, and emotional factors

Answer 224

* Fails to capture causal influences and non-quantitative factors * Fails to model decision shortcuts, psychological biases, etc. * Doesn’t capture life events (e.g. marriage) that trigger a life insurance purchase

Answer 225

* Can operate on variables about the individual (age, education, etc.) * Can predict behaviors triggered by environmental changes

Answer 226

* Move from statistical averages to understanding household decision-making * Model the causal structure of decision making, including the effect of biases

Answer 227

* “Smarter” databases containing data “micro data” * Exponential advances in computational power * Better understanding of decision-making processes and behaviors

Answer 228

* Lapse-supported if * PV profits under test \< PV profits with normal lapse assumption * Else, the product is unlikely to ever be lapse-supported * ASOP 24: can’t show illustrations if policy fails self support test

Answer 229

* guaranteed level premiums * low/zero CSV * high late benefits

Answer 230

* results in insufficient reserves * warning sign: higher-than-expected early profits

Answer 231

* ultimate lapses, not early lapses * lower-than-expected early lapses increase profit * lower-than-expected ultimate lapses increase benefit costs

Answer 232

* High income market * Competitive/marketing pressures keep premiums low * Policyholders are least likely to lapse * Agent persistency data may be an illusion due to past replacement activity

Answer 233

* Less risky if lapses are driven by independent events (e.g. employee turnover) * Help blocks/companies with poor persistency * Increase profitability in low income markets

Answer 234

* ROP riders create a tontine – everyone pays in, but only “survivors” collect ROP * Bad investment for policyholder on lapse-adjusted basis * ROP cash value is not enough to lower the risk of low lapses * ROP can convert limited-pay products to lapse-supported

Answer 235

* Many companies simply assume 100% premium persistency * Aggregate premium falls with duration due to limited pay business * Companies often vary persistency by duration

Answer 236

* **Pricing assumes more patterns in general** * Some companies use a weighted average persistency assumption * Companies adjust ULSG pricing persistency to prevent lapses

Answer 237

* Dynamic premium persistency assumptions * Sensitivity testing \*few companies currently doing

Answer 238

* when an insurer buys a life insurance policy and then recieves the death benefit when the insured dies * the first life settlements that showed up in the 1980s (focused on life expectancies \< 2 years)

Answer 239

* High growth market * Popular with large face policies owned by impaired lives * Policyholders sell their policies to the highest bidder * Policyholder receives more than CSV * Investor holds policy at fair market value

Answer 240

* Investors taking more risk—buying policies with longer life expectancies * Overpayment due to intense bidding * Lack of compensation disclosure

Answer 241

1. Insured Borrows Money To Buy Policy 2. Charity-Owned Life Insurance 3. Stranger-Owned Life Insurance

Answer 242

* Works like charity-owned

Answer 243

* Investors lend money to purchase a policy * The charity has an insurable interest in the donor * When insured dies, investors and charity split the DB

Answer 244

Wait 2 years, then policyholder can. . . * Repay loan and keep policy or * Transfer policy to lender as a life settlement

Answer 245

* Pay policyholders a higher CSV based on updated underwriting

Answer 246

1. Limit number of illustrations provided and response time 2. Public announcements portraying settlements as a bad deal for policyholders 3. “If you can’t beat ’em, join ’em” **Insurers mostly don’t like life settlements**

Answer 247

1. Continued strong demand for larger face policies 2. Premium-financed policies transfered to investors 3. Settlement companies’ ability to estimate life expectancy more accurately than standard tables 4. Preferred underwriting may wear off =\> settlement value \> cash value 5. Investors will likely target low premium products to exploit mispricing

Answer 248

* Settlement policies will get minimum funding needed to keep them in force * High persistency and high mortality, especially at older ages * Inefficiencies in policyholder behavior may go away * AXXX reinsurance is getting harder to find * Economic underwriting – underwriting may ultimately improve thanks to investors * High settlement activity can be a sign that the product is mispriced

Answer 249

* LE falls with age and is lower for males * Most life settlements involve impaired insureds (lower LE than standard)

Answer 250

* difference falls with age * difference has been falling overall with time

Answer 251

* Underwriting is getting better and/or more conservative * LEs may be improving * Increased investor demand for longer LEs

Answer 252

**Typical life settlement: $1–10MM UL policy on a NS Male age 70-something** * Mostly UL (88%), then VUL, WL, and term * Most owned by trusts or individuals * **Standard life settlement** (most common) – medically underwritten (DB\>$1MM) * **Simplified life settlement** – less underwriting, smaller policies (DB * Most involve single lives (3 out of 4 are male) * Nearly all are non-smoker * Insured age distribution is symmetric around age 75

Answer 253

**Retained DB has been increasing over time** * Benefits for investors * Better aligns incentives of investors and policyholders * Death is reported more promptly * Benefits for policyholders * Retain some DB while eliminating premium payments * Appealing to policyholders who can’t afford premiums or new coverage

Answer 254

* Close faster (less underwriting) * Market is spreading to middle class * Baby boomers need the cash for other things

Answer 255

Expected IRR = discount rate such that: PV(DB Payable to Investor) - PV(Premiums) = Purchase Price \*IRR can also be expressed as a spread over treasuries ILA

Answer 256

* 12.5% * IRRs are positive even if actual LE is 3 years higher than expected

Answer 257

* H0: Retained DB has **no effect** on expected IRR * H1: Higher retained DB ⇒ **lower** expected IRR * H2: Higher retained DB ⇒ **higher** expected IRR

Answer 258

* “Continued demand hypothesis” * Policyholders are willing to accept lower offers to retain more DB * Lower price ⇒ higher IRR for investor

Answer 259

* Adverse selection: unhealthy insureds should want to retain more DB * Investors are more likely to overpay as retained DB decreases (“single-crossing property”) * Higher price ⇒ lower IRR for investor

Answer 260

* H3: **Higher** premium convexity ⇒ **higher** IRR * Increasing premium schedules create risk for investor * H4: **Smaller** policies create diversification ⇒ **lower** IRR * Buying larger numbers of small policies creates diversification * H5: **Large** policies dilute fixed costs ⇒ **lower** IRR * Cost per amount of DB is lower for high DB policies * H6: **More** underwriting LE estimates ⇒ **lower** IRR * More information on life expectancy reduces investor risk ## Footnote **H4 & H5 oppose each other!**

Answer 261

* ß Value = 0.00 to 0.03 * Interpretation: \>10 million face policies have up to 3% higher IRR than \< 1 million face policies

Answer 262

* ß Value = -0.0055, -0.0044 * Interpretation: IRR is lower if have more than 1 LE estimate (supports H6)

Answer 263

* ß Value = -0.04 * Interpretation: Simplifed settlements for lower face policies have lower IRRs (supports H4, not H5)

Answer 264

* ß Value = 0.02 * Interpretation: IRR increases as premium slope increases (supports H3)

Answer 265

* ß Value = 0.11 to 0.14 * Interpretation: Presence of retained DBs increases IRR (supports H2, not H1)

Answer 266

* ß Value = 0.00 to 0.01 * Interpretation: 1–10 million face policies have up to 1% higher IRR than

Answer 267

* The results show that expected life settlement returns are **driven by economic factors**: * H2: Continued demand for life insurance hypothesis * H3: Premium convexity hypothesis * H4: Diversification hypothesis * H6: Model uncertainty hypothesis * Expected IRRs are **not driven by adverse selection** by policyholders

Answer 268

* **GLWB and GMIB are most popular by far** * 88% of VA contracts in study had a GLWB or GMIB * GLWB has the highest sales growth (most premium occurs year 1) * The GLWB is significantly more popular than the non-lifetime version (GMWB) * **GLWB and GMIB have the _highest persistency_ (very low surrender rates)** * GMAB is popular with younger buyers

Answer 269

1. When withdrawals start 2. Method of withdrawals 3. Amount of withdrawal taken 4. Surrender rates

Answer 270

* factors that result in **_low surrenders_**: * Older owners (65+) in general—even lower if taking withdrawals * Younger owners not taking withdrawals yet * SWPs have lower surrenders than non-systematic * Withdrawal amount ≈ 100% of max amount (not too far below/above 100%) * Higher ITM ⇒ lower surrenders * Still in surrender charge period (higher lapse when exiting)

Answer 271

* Most owners take the max allowed by contract (e.g. 4% of benefit base per year) * Younger owners are more likely to exceed max * ITM has little to no effect

Answer 272

* SWPs most significant * SWPs are popular for lifetime benefits and older owners (65+) * Withdrawals usually continue once they start (low surrenders)

Answer 273

* age and source of funding most significant * after age 70, withdrawals increase with age and duration * 2/3 of contracts are funded with _qualified money_ (tax-favored retirement funds)

Answer 274

* **Expect high utilization of lifetime benefits** (hence low lapses) * Many will be used to fund retirement * GLB riders are popular * **Qualified money has the highest withdrawal risk** * **Withdrawal activity can predict surrender activity** * **Modeling GLB utilization is important, especially for lifetime options**

Answer 275

* All offer step-up options (i.e. ratchet to AV) * Differences with GLWB * No automatic increase to benefit base if delay withdrawals * No cap on benefit base

Answer 276

* Most common benefit base design: 100% of premiums paid in * Typical waiting period: 10+ years

Answer 277

* Most common design: roll-up or max(roll-up, ratchet) * Most are B-share (have SCs) * Annuitization rates are extremely low (0.3%), but increase with age, size, and ITM

Answer 278

* Lifetime benefits available from ages 50–99 (usually 54–84) * Median max withdrawal allowed = 4% * Benefit base usually reduced proportionally for excess withdrawals

Answer 279

* **IN Scope: applies to actuaries who price life and annuity products** * Products written on individual policy forms or equivalent * **NOT in scope:** * Changes to non-guaranteed elements or dividends on in-force policies * See ASOP 2 * Reinsurance contract pricing * Illustration of non-guaranteed charges or benefits

Answer 280

* Intended **product design** objectives * Intended **market**, anticipated sales, and competitive alternatives * How the product will be **sold** (underwriting, distribution, etc.) * How the product will be **administered** (valuation systems, operational risks, etc.) * **Risk mitigation** strategies (reinsurance, hedging, etc.) * Applicable **law** (statutes, regulations, etc.) * **Tax** treatment of the product

Answer 281

* Choice of **profitability metrics** * **Targets** for profitability metrics (product level, cell level, etc.) * Approach for reflecting **risk capital** * Relevant **risk management policies** (e.g. earnings volatility limits)

Answer 282

* actuary should consider more than one (IRR, ROE, BEY, etc.) * Considerations When Selecting a Profitability Metric * Expected **pattern of profits** over time * Significance of **underlying risks** (e.g. size and pattern of risk capital) * Any **other** relevant considerations

Answer 283

* Time horizon * Granularity * Dynamic assumptions * Asset returns * Economic scenarios (MC or RW) * Accounting and actuarial bases * Risk capital framework * Taxes * Risk evaluation * Risk mitigation * Model validation

Answer 284

* sufficient to capture profitability and risks

Answer 285

* include necessary detail (time intervals, assumptions, etc.)

Answer 286

* policyholder behavior, etc.

Answer 287

* should be consistent with company practice

Answer 288

* (MC or RW) – represent an appropriate range

Answer 289

* statutory, GAAP, tax, etc.

Answer 290

* reflect the framework used in practice

Answer 291

* model should be based on expected tax structure and related items

Answer 292

* model should evaluate risks appropriately

Answer 293

* reflect strategies that will support the product

Answer 294

* model should be sufficiently transparent

Answer 295

* based on **relevant and credible** data * **margins** (credibility, uncertainty, appropriateness in aggregate) * **consistent internally**, with company practices, and with similar assumptions * reflect **interdependencies** with each other * **document** assumptions, margins, rationale, and any data modifications

Answer 296

* Anticipated sales distribution * RW or MC theory for investment/economic market assumptions * Mortality/morbidity * Policyholder behavior (elective benefits, lapses, etc.) * Anticipated expense inflation and expense allocation basis * In-force management strategies (e.g. policyholder dividends) * Capital market assumptions – replicating portfolio performance

Answer 297

* **Sensitivity Analysis** – evaluate how assumption deviations impact profitability (more for more impactful assumptions) * **Stochastic Analysis** – appropriate for interest rate and equity returns * Consider **risk mitigation strategies**

Answer 298

\*Consider and document any used in pricing * Effective **oversight** of methods and assumptions * **Protection** of the model from unintentional/untested changes * **Model input validation** * **Consistency** of model values with those calculated outside the model * **Validation of projected model values** * **Review** of assumptions by a knowledgeable person

Answer 299

Required disclosures: * **Criteria** of the actuary’s principal * **Relevant product characteristics** * **Profitability metrics** * **Considerations** used to determine the model * **Material pricing assumptions** * **Results of risk evaluation** * **Reliance** on * Governance and controls used by others * Data or other information supplied by others * Assumptions provided by others * Results of the **profitability analysis**

Answer 300

* actuarial work with an information date \>= 12 months after this ASOP is adopted by the ASB

Answer 301

* **ALL practice areas** * Developing, selecting/choosing assumptions, analysis of data/experience, * industry studies, trends, economic forecasts, etc. * Includes the selection of assumption methodology * Specific practice ASOPs govern if they conflict with this ASOP

Answer 302

* Provides guidance to actuaries performing actuarial services that include * Setting assumptions * Assessing the reasonableness of assumptions set by others

Answer 303

* Nature of assignment * Availability/relevance/credibility of data * Other available/relevant information * Will future experience differ significantly from past experience?

Answer 304

* Consider reasonableness of **material assumptions**, selection methodology, and any tendency to misestimate * Assess whether assumptions are **reasonable in aggregate** (determine adjustments) * Ensure assumptions are NOT set to **counteract** prescribed assumptions * **Disclose material inconsistencies** in non-prescribed assumptions

Answer 305

* **Margins for Adverse Deviations** – consider appropriateness * Consider uncertainty in underlying data * **Changes in Conditions Through the Information Date:** * _Internal circumstances_: changes in claims processing, business mix, etc. * _External circumstances_: changes in economy, laws, regulations, technology, etc.)

Answer 306

* disclose but don’t assess

Answer 307

* assess reasonableness * follow Precept 8 of the Code of Professional Conduct (don’t mislead!)

Answer 308

* Disclose reliance on experts (investment advisers, economists, etc.) * Actual assumption MUST be set based on the actuary’s judgment * If actuaries collaborate, the signing actuary is responsible

Answer 309

* Evaluate the alternative assumptions using sensitivity analysis

Answer 310

**Disclose the following if practical and relevant:** * **Material assumptions** – provide sufficient detail for another qualified actuary to assess * **Material changes** in assumptions since the most recent comparable results * **Events on or after the information date** that may affect the assumptions * **Material inconsistencies in assumptions** along with reasons for * **What the actuary is taking responsibility for** (setting the assumptions) Refer to ASOP 41: “Actuarial Communications” for additional guidance

Answer 311

1. **Lack of past experience** consistent with today’s UW criteria 2. Uncertainty about: * **Preferred/residual ratios**—how they change over time * **Slope of aggregate rates** over S&U period * **Anti-selection** after level term period unknown 3. Mortality improvement * Lack of **consistent** data * **Uncertainty** whether past improvement will continue in future

Answer 312

* n% of reverters are fully select

Answer 313

* all reverters’ mortality = fully select + extra amount that grades off

Answer 314

* Both take a conservation of deaths approach * Both assume the excess lapse group is made up of 1. Select lives 2. Normal lives * Effect of both: **Lower reverter mortality** =\> **higher persister mortality**

Answer 315

* Vary total and excess lapse rates assumed * Vary parameters in D-M and B-K approaches * Assume no profits after level premium period

Answer 316

* Loss of profit from high premiums paid by persisters * Lower GAAP income * Not as marketable to policyholders

Answer 317

* No classification by underwriting class * Ignores premium size in ART period * Assumes 100% rational policyholder behavior

Answer 318

* Aggregate **duration 10**, 11, and 12 lapses = **70%**, 40%, \< 20% * Shock lapses **increase** with: * Premium jump size * Issue age * Males * Super preferred * Face amount * Annual mode (lower frequencies) * Issue age and/or face within premium jump group * Shock lapses are heavily skewed to **EOY 10 and BOY 11**

Answer 319

* Mortality deterioration is highly correlated with shock lapses * Rises at an **exponential** rate with shock lapse size

Answer 320

* **Aggregate duration 11 mortality = 300% of level period** * Drops to 200% in duration 12, then declines slowly * **Mortality _increases_ with the same factors that increase shock lapse:** * Premium jump size * Issue age * Males (slightly) * Super preferred * Face amount * **Mortality deterioration is skewed toward BOY 11**

Answer 321

* Overall: experience is very **limited** * Degree of **mortality deterioration** is unknown * Early lapse experience has been **lower** than expected

Answer 322

* Policyholder complacency (automatic bank draft) * Still shopping for lower rate * Some may be OK with higher cost * Reluctance to go through underwriting * Personal situations that delay lapse (e.g. divorce)

Answer 323

* **The Approach**: Assume shock lapses and develop a single YRT ceiling * **Advantage**: simplest to administer * **Disadvantages** * Worst risks remain in force * Claims volatility due to limited credibility * Potential for negative publicity

Answer 324

* **The Approach**: * Optional questionnaire determines PLT rate class (e.g. SM/NS) * If decline survey: default to original YRT ceiling

Answer 325

* **Advantages**: * Greater sense of fairness to policyholders and regulators * Lowers selective lapsation * Insured gets lower rate and insurer has more confidence in rates

Answer 326

* **Disadvantages:** * May be an early signal that increases lapses and/or term conversions * Must solve implementation problems * Communication to policyholders, questionnaire design, how to handle incomplete questionnaires, how to maximize response rate

Answer 327

* **The Approach:** * PLT premiums increase at a smaller increment intially (e.g. 5 years) before reaching YRT scale

Answer 328

* **Advantages:** * Makes initial PLT rates more attractive (lower lapses) * Insurer can still increase to YRT ceiling * Avoids underwriting * Low administrative cost * Early experience suggests it works as intended

Answer 329

* **Disadvantages:** * Best risks can still lapse to get better rates * No reliable experience for post-graded YRT period * Experience is Canadian ⇒ will the U.S. be the same?

Answer 330

* **The Approach**: * Maintain level term class structure into PLT period * Develop separate YRT scales by class * All YRT scales converge to an ultimate scale (original YRT scale)

Answer 331

* **Advantages**: * May be “the fairest approach” since it uses original underwriting * YRT scale encourages/discourages lapses based on class * Permanent insurance experience could be used for classes

Answer 332

* **Disadvantages:** * Lack of experience with the approach * Preferred classes will have lowest initial YRT premiums but steepest slope * Permanent pricing can’t be used directly * Selective lapse risk from preferred insureds who have become impaired

Answer 333

* **Shock lapse** – high, additional lapse rate at end of level premium period * Healthy lives aren’t willing to pay the higher post-level premiums

Answer 334

* **D-M predicts the expected mortality after a shock lapse** * Split shock lapses into and persisters and reverters: * **Persisters** – lives remaining after the shock lapse * **Reverters** – lives in the shock lapse group * **Effectiveness rate** – The percentage of reverters that qualify for select mortality * Use the **conservation of deaths principle** to solve for persister mortality

Answer 335

q^norm = q^sel x w^eff + q^norm x (w^non-eff + w^base) + q^pers x (1 - w) q^sel= select issue age mortality q^norm= normal, point-in-scale mortality q^pers= mortality of persisters remaining in force w = total lapse rate at the time of the shock lapse w^base= normal base lapse rate in absence of shock lapse w^shock= w - w^base w^eff = "effective reverters" = effectiveness rate x w^shock w^non-eff = non-effective reverters = w^shock - w^eff

Answer 336

* **Shock lapses and expected mortality are closely related** * Do not set assumptions independently! * **D-M mortality predictions are sensitive to the effectiveness rate assumed** * **Effectiveness depends on policyholder behavior** * Information about health * Ability to replace coverage * Other factors also drive shock lapses, but **D-M provides useful boundaries**

Answer 337

* Ultimate lapse rate **falls** as term length **increases**: * T10: 6% * T15: 4% * T20: 3%

Answer 338

* *Lower lapses for. . .** 1. Older issue ages 2. Preferred classes 3. Non-smokers 4. Joint products Male lapses ≈ female lapses

Answer 339

* **T100: Guaranteed Level Premiums to 100, no CSV** * Early pricing assumed 6% ultimate lapse rates * Actual ultimate lapse rates were less than 2% (oops!) * Lower lapses led to much **higher reserves** and **lower income** * Lessons learned can apply to other lapse supported products * Examples include SGUL, LTCI, ROP Term

Answer 340

* Ceding company gains if mortality \> expected * Possible if reinsurance premiums \> T100 premium

Answer 341

* Term conversions have a higher mortality cost due to anti-selection: q^c_{[x] + t} \> q^s_{x + t} * Companies may not even know the cost of their term conversions * Data credibility problems: administrative system data is poor or doesn’t exist * Historical data may **understate** conversion activity

Answer 342

* Use industry data to validate company-specific data * Reinsurers can help direct writers by sharing data and insight

Answer 343

* Term products are VERY competitive * Sales are VERY sensitive to price * LOTS of pressure to keep premiums LOW * Conversions add value but also cost → where do you reflect the cost?

Answer 344

* **Features that restrict conversions** 1. Limit to N years after issue 2. Maximum attained age 3. Not available during active disability premium waiver * **Products that encourage conversions** * ART-to-permanent conversions (targeted at younger buyers)

Answer 345

* **Include in permanent product pricing** * **Advantage**: keeps term premium low * **Difficulties**: hard to estimate conversion utilization and permanent plan sales volume * **Solutions**: * Use conservative load in permanent premium * Develop a conversions-only product (may be unattractive) * **Include in term pricing** * **Advantage**: aligns conversion cost with originating term product * Include cost of agent compensation and conversion credits * **Include in term pricing but scale back availability** * Limit to N years after issue * Offer 2 term products

Answer 346

* Use **point-in-scale** YRT rates * **Coinsurance** treaties: YRT rates apply only to conversions * **YRT** treaties: * Increase overall YRT rates OR * Use separate YRT rates for conversions

Answer 347

* May already be covered if also reinsure term plans * Future future conversions, 3 areas of uncertainty: mortality, volume, mix * Solutions: * Separate YRT scale for converted policies * Single, substantially loaded YRT scale

Answer 348

* First, calculate the following A/E ratios: * Converted A/E * Level Period A/E * Can be count- or amount-based * PISM ratios = ratios of the above A/E ratios: Converted A/E / Level Period A/E * level period experience is a proxy for true permanent policy experience

Answer 349

* **Term conversion rates are highest at end of level period** * Late duration conversion rate ≈ 5x first year * Under 1% in LP, then jump to 5% late in LP * Larger policies are more likely to convert

Answer 350

* **Post-conversion mortality is much higher for late duration converters** * PISM ratios are 180–220% in the first year after conversion * Mortality is higher for large policies (PISM by amount \> PISM by count) * PISM ratios decline but stay above 100% (anti-selection reduces)

Answer 351

* Highest in the initial durations after conversion (4–6%) * Late converters have lower lapse rates than early or mid

Answer 352

* **30–60 day period after premium due date** * Policyholder can still make past due payment without lapsing * Without adjustments, lapses will spike in month 2 of each policy year * Adjustment moves the lapse back to premium due date * Results in higher proportion of lapses in month 12 of each policy year * Same concept applies to conversions

Answer 353

1. Reflect conversion cost in term pricing 2. Develop separate permanent premiums for conversions 3. Reflect extra conversion mortality in permanent policy pricing

Answer 354

* Might be appealing if insurer has mature experience * Can make permanent mortality artificially high

Answer 355

* Avoids cross-subsidization but is administratively challenging * Alternatively, could reinsure conversions using YRT PISM rates

Answer 356

* (most common) * This is what the author of this article advocates * Allows reserves to build for conversions * Aligns revenue with the source of risk (term “swaption”)

Answer 357

1. At issue, determine the expected extra mortality cost for term conversions 2. Calculate the charge for term policyholders

Answer 358

For each year when a conversion is possible calculate: Extra NSP for Conversions = K_x,r = PVFB^Converted- PVFB^{Standard WL} = reserve for conversions at duration r

Answer 359

A_x,r = _rp_x x e_x,r x K_x,r x v^r Total PV of Expected excess mortality = Σ A_x,r Level Conversion Charge = (Total PVFB of Excess Mortality / ä_x:n)

Answer 360

* **Conversion cost increases when base mortality increases** * Higher issue ages * Males (compared to females) * Smokers (compared to non-smokers) * Non-preferred (compared to preferred classes)

Answer 361

* **Restricting** the conversion period **reduces** the cost of conversion * Disallowing conversions after year 10, reduces cost roughly 30% * Caveat: policyholders will accelerate conversions in response