Session 48 PD, Mortality Update. Moderator: James M. Filmore, FSA, MAAA

Transcription

1 Session 48 PD, Mortality Update Moderator: James M. Filmore, FSA, MAAA Presenters: Thomas P. Edwalds, FSA, ACAS, MAAA Dieter S. Gaubatz, FSA, FCIA, MAAA

2 2015 VBT Table Development Tom Edwalds, FSA, ACAS, MAAA DePaul University

3 2015 VBT Acknowledgement Based on presentation Mary Bahna Nolan, chair : AAA Life Experience Committee SoA Preferred Mortality Project Oversight Group Originally presented to NAIC Life Actuarial Task Force on August 14, 2014 Uses that presentation with permission and provides additional details and updates

4 2015 VBT Table Development Preliminary data segment analysis Ultimate unismoke experience table Select experience table Smoker distinct ultimate table 2015 VBT Primary tables 2015 VBT RR tables

5 Preliminary Data Segment Analysis Advanced ages Maximum q x defined as.5 No omega age No grading to population data Juvenile issue ages Select period definition

6 Juvenile Issue Ages Consider issue ages 0 17 as juvenile policies Examined mortality relative to population mortality & insured mortality (2008 VBT) No clear relationship to population mortality No smoker/non smoker distinction No observable select period All durations for juvenile issue ages considered ultimate NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

7 Graduation Approach Split data into select dataset & ultimate dataset Used Generalized Additive Models (GAM) Created 2 models mortality rates amount: 1. Unismoke ultimate model 2. Smoker distinct select model Both models used essentially all relevant available data Total amount excluded was trivial NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

8 Ultimate Data GAM Model Excluded exposures & claims for attained ages > 105 Significant mortality predictors in dataset: Gender Issue year era Attained age Face amount band Issue age Decided not to include era & band in final model Overwhelming majority ultimate data from pre 1980 issue era, face amounts < $10K Also interaction issue year era & face amount band NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

9 Ultimate Data Issue Age Effect Issue age effect primarily due to difference between juvenile and adult issue ages Therefore, 2 separate GAM submodels were fit: Juvenile issue ages 0 17 (all durations) Attained age 0 excluded, handled separately Adult issue ages 18+ Durations 26+ for males at issue ages Durations 21+ for females at issue ages At higher issue ages, more durations included NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

10 Ultimate Unismoke Model Construction Age 0: raw data Ages 1 35: juvenile issue age GAM model Ages 35 45: log linear interpolation Ages 45 95: adult issue age GAM model Ages : cubic extrapolation Ages 112+: q(x) =.5

11

12 Select Data GAM Model Excluded exposures & claims at issue ages > 90 Significant mortality predictors in dataset: Gender Smoker status Issue age Duration Issue year era Face amount band Issue year era & face amount band not included in final model Too complex to model & present in mortality tables 96 tables before split into RR levels Ultimate bands not same as select bands NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

13 Monotonicity Constraints 1. Vertical constraint: For same gender & smoker status, attained age > Horizontal constraint: For same gender & smoker status, attained age > Diagonal constraint: For same gender & smoker status 4. Gender constraint: For same smoker status 5. Smoker constraint: For same gender NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

14 Select Data GAM Model Adjustments Raw GAM model inspected for monotonicity violations Horizontal, vertical, & diagonal constraints first Model adjusted to meet constraints using various techniques: Linear interpolation between adjacent rates Linear interpolation between selection wearf patterns adjacent rates Fitting smooth selection wearf patterns e.g., quadratic arcs along attained age diagonals Manual adjustments pure judgment Most adjustments < +/ 5%

15 Select Data GAM Model Adjustments Gender constraint: No adjustment to male submodels Female q s capped at 98% male q s Smoker constraint: No furr adjustment to nonsmoker submodels Smoker q s floored at nonsmoker q s Rare violations eir constraint At edges model

16 Furr Adjustments to Select Model Adjustments to young adult issue ages Crude select model q s for male young adult issue ages seemed too low compared to raw q s Smooth sets adjustment factors developed Male NS, issue ages 18 31, durations 1 15 Male SM, issue ages 29 36, durations 1 7 Older issue age q s debated extensively NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

17 Older Issue Age Investigation Industry feedback 08 VBT q s for older issue ages too high 2014 VBT Team investigation Careful review ILEC raw data & model level & slope older issue age q s Reviewed MIMSA & TOAMS older age data Overlap w ILEC data, but sufficient variation Determined that GAM model needed older age adjustment NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

18 Older Age Adjustment to GAM Model Adjustment factors developed & applied for: Male NS, issue ages 70 90, durations 1 10 Male SM, issue ages 61 81, durations 5 14 Final rates provide reasonable balance between: Raw ILEC older issue age experience data Prior estimates older issue age mortality Need for smooth transition from select to ultimate Relatively small number claims underlying raw experience data NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

19 Ultimate Smoker Distinct Model Smoker distinct ultimate data very thin Used smoker distinct select model to split unismoke ultimate model smoking status Split ultimate data presumed smoking status using same factors NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

20 Split Ultimate Unismoke Model to Smoker Distinct 1. Extrapolated smoker distinct late duration select q s to 1st ultimate duration 2. Computed implied ultimate SM/NS mortality ratio By attained age and gender 3. Smood SM/NS ratios at high and low ages Graded to 1 at age NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

21 Split Ultimate Unismoke Model to Smoker Distinct 4. Computed implied smoker prevalence rates: Compared extrapolated smoker distinct ultimate q s to unismoke ultimate model Algebraically solved for implied nonsmoker prevalence Smood prevalence at high ages 5. Applied smoker prevalence & SM/NS ratios to unismoke ultimate model 6. Applied same ratios to raw experience Presumed smoker distinct data, for graphing NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

22 02 09 Table Sample Graphs

23 02 09 Table Sample Graphs

24 02 09 Table Sample Graphs

25 02 09 Table Sample Graphs

26 2015 VBT Table Construction Applied factors to experience table Improvement factors to 1/1/2014, crafted 2014 VBT Added 1 more year improvement to 1/1/2015 Re applied monotonicity constraints Developed RR tables Duration 1 relativity factors Used same factors as 08 VBT for same RR level Interpolated or extrapolated for new RR levels Used same wearf factors as 08 VBT Re applied monotonicity constraints

27 VBT Select Graphs

28 VBT Select Graphs

29 VBT Select Graphs

30 VBT Select Graphs

31 Table Comparisons 20 year period Discount rate 3.5% Assumes deaths occur at end each year and no lapses Provided Suzanne Chapa, FSA

32 Table Comparisons 20 year period Discount rate 3.5% Assumes deaths occur at end each year and no lapses Provided Suzanne Chapa, FSA

33 Table Comparisons 20 year period Discount rate 3.5% Assumes deaths occur at end each year and no lapses Provided Suzanne Chapa, FSA

34 Table Comparisons 20 year period Discount rate 3.5% Assumes deaths occur at end each year and no lapses Provided Suzanne Chapa, FSA

35 Questions?

36 MORTALITY UPDATE SESSION 48 SOA L&A SYMPOSIUM; MAY 5, 2015 Dieter Gaubatz, FSA, FCIA, MAAA

37 Table contents 1. SOA recent mortality activities VBT CSO 2

38 SOA mortality studies Annuities Pensions Uninsured private pension plans Nov New table RP ,000 claims Observation years Increased funding requirements for a number plans Additional Comments report Group Annuities Group pension contracts issued insurance carriers March ,000 claims Observation years Improvement increased 0.7% / yr faster than scale AA Slow down extra improvement from prior study from OYs: was 2% per year faster Individual Payout Annuities Immediate annuities, annuitizations and life settlement options July 2014, with info added Jan ,000 claims Observation years Annual improvement 0.8% for males, 1.1% for females since prior study 3

39 SOA mortality studies Group Life Life report Sep 2014 latest version 2014/12 addendum Improvement rates between 2006 and 2013 reports Ann l improvements age and gender (-1% to 4%) Waiver incidence rates 4

40 2015 VBT / 2017 CSO Acknowledgement Mary Bahna-Nolan, chair Life Experience Committee / Society Project Oversight Group Heavily used her presentation to NAIC Life Actuarial Task Force, with permission This presentation uses that presentation as a base and provides additional updates which have since occurred 5

41 Status to-date 1. NAIC meeting March 26, VBT table Approved exposing VBT and VBT RR tables for 45 days; Projection one more year from 2014 to 2015 Exposed late April, CSO table Target approval at NAIC August, 2015 meeting at latest 6

42 2015 VBT table package 2015 VBT Structure NS/SM/Composite ALB/ANB Age, gender, select and ultimate Juvenile rates unismoke only and no select period RR NS 50, 60, 70, 80, 90, 100, 110, 125, 150, 175 RR SM 75, 100, 125, VBT RR tables NS 70, 80, 90, 100, 110, 120, 130, 140, 150, 160 SM 75, 100, 125, 150 Omega rate per K, beginning at age per K, beginning at age 110 No omega age No limited underwriting table A separate project; review after GI/SI tables are completed 7

43 2017 CSO table package 2017 CSO CSO Tables No RR tables CSO Preferred Structure Tables 3 Non-smoker 2 - Smoker Omega age for CSO table is 121 q 121 = 1000 per K 8

44 Table uses Tables 2015 VBT Basic Tables (RR100) Purpose AG38 VM 20 Deterministic reserves 2015 VBT Relative Risk Tables VM 20 Stochastic reserves 2017 CSO and 2017 CSO Preferred Structure Tables PBR Margins Net premium reserves Tax reserves Non forfeiture determination Basis for 7702/7702A Cap for universal life cost insurance charges VM 20 Deterministic and Stochastic reserves NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

45 Underlying experience The 2015 VBT tables are based on industry experience, a significant increase in exposure and number claims over Study/Table By Amount By Number Number Number studies underlying both 2008 and 2001 VBT Claims table development. Exposure Actual deaths Companies / 2015 VBT $30.7 trillion 266 million 2.5 million / 2008 VBT $7.4 trillion 75 million 0.7 million / 2001 VBT $5.7 trillion 175 million 1.25 million 21 Increase from 2008 VBT +315% +255% +257% +46% Increase from 2001 VBT +439% +52% +100% +143% NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

46 Underlying experience, cont d Or significant increases in 2008 VBT experience: Face amount exposure years ($ trillion) Characteristic Increase No. exposure years % Female % Issue Age % Attained Age % Smoker % Also: More tobacco definition (vs. cigarette smoker) Preferred NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

47 Underlying experience, cont d Expected basis 2008 VBT primary (RR 100) table Overall, mortality improvement from data underlying 2008 VBT Study Period Male Female Aggregate Exposure ($ Trillion) #Death Claims (underlying 2008 VBT) 101.1% 100.5% 100.9% , (underlying 2015 VBT) 94.4% 94.9% 94.5% ,559, experience for common companies to study 92.3% 94.3% 92.8% ,940, k+ 88.5% 89.4% 88.5% , k+ 84.2% 85.7% 84.5% ,634 Annual improvement rate All 1.11%, common 1.44% Note: not underlying annual mortality improvement NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

48 Underlying experience, cont d Life insurance mortality experience varied many factors including face amount, smoker status, and issue age. A/E* Ratio NS versus SM Smoker Status A/E Ratio Amount Non smoker 92.5% Smoker 97.7% Unknown Status 100.1% A/E* Ratio By Issue Age Issue Age A/E Ratio Amount % % 80 89** 61.6% A/E* Ratio By Amount Face Amount Band ($) A/E Ratio Amount 50,000 99, % 100, , % 250, , % 1,000,000 2,499, % 2,500,000 4,999, % 5,000,000 9,999, % 10,000, % Aggregate 94.3% * Expected basis = 2008 VBT Primary Tables ** for common companies drops to 55% Source: Society, compilation Individual Life Experience Reports 2002 through NAIC Life Actuarial Task Force Meeting 23 The Year in Review, November August 14,

49 Underlying experience, cont d Actual to Expected (A/E) comparison, cont d A/E Ratios for contributing companies non smoker risks By amount Average 92.5% Range [36% 1,164%] By count Average 110.1% Range [49% 863%] Expected basis = 2008 VBT RR 100 Table NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

50 Underlying experience, cont d Actual to Expected (A/E) comparison, cont d A/E Ratios for contributing companies Smoker risks By amount Average 97.7% Range [41% 194%] By count Average 110.0% Range [75% 184%] Expected basis = 2008 VBT RR 100 Table NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

51 Select period Select Period Issue Age MALE FEMALE Issue Age MALE FEMALE Source: 2014 Valuation Basic Table Team Society & Joint Project Oversight Group NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

52 Select period Selection wear-f formulas Relative rate at duration 1 = q [x]+0 / q x At duration t = q [x]+t-1 / q x+t Select / ultimate rates using same attained age Rate difference wear-f at duration t = 1 - (1 q [x]+t-1 / q x+t-1 ) / (1 q [x+t-1]+0 / q x+t-1 ) 17

53 Select period Selection wear-f Relative rate Selection wear-f Duration (t) Issue Age (x) Male non-smokers primary (RR 100) table 25* 38% 33% 49% 56% 69% 35* 20% 25% 36% 64% 83% 45 22% 30% 56% 88% 88% 55 19% 35% 64% 84% 91% 65 26% 44% 64% 84% 100% 75 18% 45% 85% 100% 100% 85 13% 58% 100% 100% 100% * ultimate rate for attained ages up to 43 is select rate for issue age 18 Relative rate at duration 1 = 1- q [x]+0 / ult q x Selection wear-f at duration t = 1- (1 - q [x]+t-1 / ult q x+25 )/ (1 - q [x[+0 / ult q x+25 ) 18

54 Select period Varies issue age and gender Considered both observable as well as prospective select period Underlying select period is independent preferred wear f Observable select period Reviewed both common and all companies Data analyzed based on count rar than amount to remove influence variations/fluctuations size claim Attempted to normalize socio economic impact over time Focused on gender/smoker status level, quinquennial attained age groupings Used GAM (Generalized Additive Model) to test fit actual mortality to mortality predicted GAM model duration; results shown as ratios to ultimate mortality at same attained age NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

55 Select period, cont d Prospective aspects Prospective expectations vs. observable past experience Impact events or changes in underwriting in underlying data Movement from unismoker to smoker/non smoker rates (1980s), Movement from smoker/non smoker to non tobacco/tobacco distinction (1990s), Liberal underwriting period with increased level underwriting exceptions ( ), Development mature age underwriting requirements such as cognitive function (2005 present) Changes in smoker prevalence, and rate change in smoking prevalence Most events thought to shorten length select period from that in observed data; a couple such as NT versus NS and older age cognitive function testing may elongate NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

56 Adjustments to underlying experience 1. Adjust data to remove post level term antiselective mortality; 2. Adjust data to recognize differences in experience from different underwriting eras; and 3. Improve underlying experience to start date table (2015) NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

57 1. Adjustment to remove effects post level term mortality Examined underlying experience for term plans only Calculated actual to expected (A/E) ratios face amount issue age group and duration in total and for 10, 15 and 20 year term plans The ratios were calculated for male and female separately and for both genders combined. No separate calculation smoker status. Recalculated A/E ratios estimating impact removing post level term (PLT) experience for each specific period term plans This provided proposed adjustment to decrease total rates to recognize PLT impact. Factors vary issue age/duration Average 2.9% at duration 13 versus 1.3% at duration NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

58 1. Adjustment to remove effects post level term mortality, cont d Adjustment factors to remove effects post level term Issue Ages Durs Durs Durs Durs % 99.3% 99.9% 99.2% % 99.6% 99.7% 97.4% % 98.8% 99.9% 98.1% % 99.3% 99.8% 98.1% % 99.2% 99.8% 99.4% % 98.4% 99.7% 100.0% % 97.1% 100.0% 100.0% % 99.1% 99.9% 100.0% % 99.6% 99.9% 100.0% % 99.8% 100.0% 100.0% % 100.0% 100.0% 100.0% % 100.0% 100.0% 100.0% % 100.0% 100.0% 100.0% % 100.0% 100.0% 100.0% NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

59 2. Select period adjustments for different underwriting eras, cont d Adjustment factors to select period mortality to account for differences in underwriting eras NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

60 3. Mortality improvement Considerations General population improvement US Vital Statistics Human Mortality Data Base (HMD) Social Security Administration Data (SSA) After looking at 3 sources, SSA data selected as source for general population Insured data Common company data for period Given short period time for historical experience and volatility from year over year, believe general population data is preferable NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

61 3. Mortality improvement, cont d Factors applied Within study period : Actual mortality improvement to adjust to end study period (2009). For period : Apply average annual improvement rates varying attained age and gender. Based on general population data (SSA) = average : (a)average annual improvement rates implied SSA s most recent intermediate level projection mortality for social security population; and (b)actual average annual improvement rates from historical SSA data for most recent 10 year period NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

62 3. Mortality improvement, cont d 2015 VBT Mortality Improvement Factors For period (and 2017) Attained Age Male Female % 0.75% % 0.75% % 0.75% % 0.75% % 0.75% % 0.75% % 0.46% % 0.17% NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

63 Relative risk (RR) tables 1. Non-smoker RR tables: needed an expansion range RR tables Keep number tables same 2. Smoker RR tables: use same range as in 2008 VBT package VBT RR table relationships used as a base to construct 2015 VBT RR tables 4. Underwriting Criteria Score (UCS) calculator revised Tool to help determine appropriate RR table for each preferred class New score is based on quantative mamatics and assumption expectations Prior version was based on qualitative pressional opinions most common criteria Online internet version will be likely be developed later in

64 Preferred wear f factors Analyzed level wear f but experience still emerging. Emerging relationships are similar to what was used in 2008 VBT analysis. The preponderance NS data being in early durations furr complicated analysis; refore, also examined Milliman s MIMSA and Towers Watson TOAMS studies. Preferred wear f factors same as for 2008 VBT, with exception that y grade f to age 95, same as underlying select period rar than 90. The extension factors used to grade from age 90 to 95 were based on pressional judgment NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

65 Preferred wear-f period formulas (issue age comparison) Relative rate at duration 1 = RR1 q [x] + 0 / RR2 q [x] For duration t: = RR1 q [x] + t -1 / RR2 q [x] + t -1 - used RR 70 as comparison table in example calcs (i.e. RR1 = 70) - base table (RR2 = 100) Preferred wear-f at duration t = 1 - (1 - RR1 q [x]+t-1 / RR2 q [x]+t-1 ) / (1 RR1 q [x]+0 / RR2 q [x]+0 ) {(1 relative rate for issue age x at duration t) / (1 relative rate for issue age x at duration 1)} 30

66 Preferred wear-f factors Comparing MNS RR 70 and RR 100 (Primary) tables Attained Age Relative rate in ultimate period ( RR70 / RR100) 2015 VBT 2008 VBT Attained Age 2015 VBT 2008 VBT 30* 78% 86% 65 71% 71% 35* 79% 85% 70 75% 75% 40* 79% 85% 75 80% 80% 45 79% 85% 80 87% 87% 50 79% 81% 85 91% 93% 55 75% 75% 90 95% 100% 60 71% 71% % 100% ultimate rate for attained ages up to 43 is select rate for issue age 18 Relative rate = 70 q x / 100 q x 31

67 Preferred wear-f factors 2015 VBT (issue age comparison) Comparing MNS RR 70 and RR 100 (primary) tables Issue Age (x) Relative Rate Preferred wear f percentages Duration (t) Attained Age 25 81% 25% 8% 15% 13% 12% % 2% 1% 3% 8% 14% % 0% 5% 12% 21% 32% % 5% 15% 27% 42% 61% % 14% 31% 51% 70% 84% % 29% 56% 81% 100% 100% % 35% 100% 100% 100% 100% 110 Complete preferred wear-f at attained age 95. Relative rate (dur 1) = 70 q [x]+0 / 100 q [x]+0 Preferred wear-f =1- (1 70 q [x]+t-1 / 100 q [x]+t-1 ) / (1 70 q [x]+0 / 100 q [x]+0 ) 32

68 Preferred wear-f factors 2008 VBT (issue age comparison) Comparing MNS RR 70 and RR 100 (primary) tables Issue Age (x) Relative Rate Preferred wear f percentages Duration (t) Attained Age 25 80% 7% 1% 1% 2% 1% % 3% 5% 1% 3% 10% % 0% 4% 10% 19% 31% % 8% 16% 28% 42% 62% % 13% 31% 51% 74% 100% % 29% 62% 100% 100% 100% % 100% 100% 100% 100% 100% 110 Complete preferred wear-f at attained age 90. Relative rate (dur 1) = 70 q [x]+0 / 100 q [x]+0 Preferred wear-f =1- (1-70 q [x]+t-1 / 100 q [x]+t-1 ) / (1-70 q [x]+0 ) / 100 q [x]+0 ) 33

69 2017 CSO Impact Study Study impact proposed tables on reserves Net premium reserves Deterministic reserve under VM-20 Results provided in two or sessions Session 55 (concurrent with this one) Session 65 (today at 10:00 11:15) 34

70 2017 CSO table package 2017 CSO CSO Tables No RR tables CSO Preferred Structure Tables 3 Non-smoker 2 - Smoker Omega age for CSO table Age 121 (q 121 = 1000 per K) 35

71 2017 CSO Table Development Margins applied NAIC LATF guidance: margins consistent with 2001 CSO to cover 70 79% mortality from contributing companies in underlying mortality study Purpose margin to cover variation an individual company s mortality around mean (company variation) NAIC Life Actuarial Task Force Meeting The Year in Review, November August 14,

72 2017 CSO preferred structure table development 1. Preferred structure tables 2015 VBT as base With different RR table structure 3 NS 2 SM Structure similar to 2001 CSO preferred tables Aggregates back to overall table 2. Relationship rates in table based on SOA ILEC experience collected for business issued and reported under a preferred structure basis Very different than 2001 preferred CSO table relative rates 37

73 Mortality Update Session 48 SOA L&A Symposium May 5,