Post on 08-Jul-2020
Session 36 PD, Practical Lessons From the Living to 100 Symposium V
Moderator: Jean-Marc Fix, FSA, MAAA
Presenters:
Jean-Marc Fix, FSA, MAAA Andrew Chong Jenkins, FSA, CERA, MAAA
Kai Kaufhold
Practical Lessons from the
Living to 100 Symposium
Jean-Marc Fix, Optimum ReAndrew Jenkins, Protective Life
Kai Kaufhold, Advanced Reinsurance Services
Session 36
• Jean-Marc: Causes of death and mortality trend pitfalls
• Kai: Modeling• Andrew: Communicating mortality information
Order
• Comparability• Independence• Extrapolation
danger
Cause of Deaths Pitfalls
Country ICD 7 to 8 ICD 8 to 9 ICD 9 to 10
USA 1968 1979 1999
Japan 1968 1979 1995
France 1968 1979 2000
Italy 1968 1979
England and Wales 1968 1979 2001
Australia 1968 1979 1998
Sweden 1969 1987 1997
Switzerland 1969 1995
Singapore 1969 1979
Norway 1969 1986 1996
Comparability
Table 1b Causes of death mortality: what do we know of their dependence, S Gaille and M Sherris
Comparability: Circulatory
Figure 2 (a) and 3(a) Causes of death mortality: …, S Gaille and M Sherris
Males USA
Comparability: Cancer
Figure 2 (b) and 3(b) Causes of death mortality: …, S Gaille and M Sherris
Males USA
Comparability: Respiratory
Figure 2 (c) and 3(c) Causes of death mortality: …, S Gaille and M Sherris
Males USA
Comparability: External Causes
Figure 2 (d) and 3(d) Causes of death mortality: …, S Gaille and M Sherris
Males USA
Comparability: Infectious and Parasitic
Figure 2 (e) and 3(e) Causes of death mortality: …, S Gaille and M Sherris
Males USA
• Dependence between causes combining to overall mortality
• Impact of changes in one component may affect different country differently
Independence
Causes of death mortality: what do we know of their dependence, S Gaille and M Sherris
• Causes of death data is sparse in life insurance• When using population data, need to look a
some confounding impacts: how much improvement is due to lower smoking prevalence?
• Obesity example: more prevalence (especially at the very young ages) but better cardiovascular mortality…
Extrapolation Danger
Obesity and mortality , Sam Gutterman
• A new angle: modal age at death
• A relook at an old angle: using what we are used to using …
Mortality Trend Pitfalls
Old wolf pit trap, Germany by Georg WassmuthPermission CC‐BY‐SA‐3.0‐DE
Mortality Rectangularization
Modal age at death: mortality trends in England and Wales 1841‐2010, Emily Clay
Mortality Compression?
Modal age at death: mortality trends in England and Wales 1841‐2010, Emily Clay
1951
1961
19711981 1991
2001
2010
Modal Age at Death
Modal age at death: mortality trends in England and Wales 1841‐2010, Emily Clay
• Age specific rate of mortality change with age kx
• kx = ln (mx)-ln(mx-1) where mx is mortality rate• If mortality is Gompertz-like then kx is constant• If kx decrease with x: deceleration of mortality
Beware your Embedded Assumptions
• Authors looked at 24 cohorts (USA, Canada, Sweden and France, for Mm & F and for 1894, 1896 and 1898
Beware your Embedded Assumptions
Mortality trajectories at extreme old ages: a comparative study of different data sources on old‐age mortalityNS Gavrilova and LA Gavrilov
• Authors looked at 24 cohorts (USA, Canada, Sweden and France, for Mm & F and for 1894, 1896 and 1898)= no significant deceleration!
• 8 cohorts exhibit non significant deceleration• But…
Beware your Embedded Assumptions
• If you combine in 5 years age group (by using kx=[ln(mx)-ln(mx-5)]/5 where mx is 5 year mortality rate
• Then 4 cohorts are now significant for deceleration and 17 other show non-significant deceleration
… or You Might Stray
Impact Estimation
Figure 8 Age‐specific rate of mortality change with age, kx, by age interval for mortality calculation. Simulated data assuming that hazard ratios follows the Gompertz Law
Mortality trajectories at extreme old ages: a comparative study of different data sources on old‐age mortalityNS Gavrilova and LA Gavrilov
• We often assume uniform distribution of death on quinquennial intervals
• At the older ages it is not even safe to do so for ONE year age interval…
One Year Time Interval
A fast computer is
nevera substitute
for an active brain
Take Home Lesson
Models and other Fun Stuff
1. Longevity is just round the bend.• Aubrey de Grey
2. Modeling advanced ages.• Gavrilova & Gavrilov, Ouellete & Bourbeau
3. Advanced experience analysis.• Zhu & Li.
4. Ideas for trading on longevity and mortality.• Li, Chan & Li.
Inspirations
• Aubrey de Grey: “Strategies for Engineered Negligible Senescence” (SENS) www.sens.org
• Aging is a lethal disease. Prevent and reverse age-related ill-health.
• Apply principles of regenerative medicine to repair the damage of aging at the cellular and molecular level.
• Postpone aging and extend healthy life-expectancy in our lifetimes!
• LESSON 1: Actuarial models may break!
Longevity is nigh!
1. Data is key for ages 80+• Howard: “Liars, cheaters and procrastinators.”• Gavrilova & Gavrilov:
US Social Security Administration Death Master File• Ouellette & Bourbeau:
French-Canadian Centenarians (parish records)
2. Distinction between death probabilities and hazards is important: 1
12
: lim→
Modeling advanced ages.
3. Mortality law for oldest ages?• Gavrilova & Gavrilov: No deceleration on a cohort basis.• Ouellette & Bourbeau: Deceleration possible.
Gompertz law (1825): →
∞
Modeling advanced ages.
Log (force of mortality): Gompertz law (1825)
Source: Richards, Kaufhold and Rosenbusch (2013)
Modeling advanced ages.
‐3.5
‐3
‐2.5
‐2
‐1.5
‐1
‐0.5
0
0.5
80 85 90 95 100
log( m
ortality ha
zard )
Age
Actual
Linear
3. Mortality law for oldest ages?• Gavrilova & Gavrilov: No deceleration• Ouellette & Bourbeau: Deceleration
Gompertz law (1825): →
∞
Thatcher-Kannisto (1998): →
1(Perks 1932)
Modeling advanced ages.
Log (force of mortality): Perks law (1932), a.k.a Thatcher-Kannisto
Source: Richards, Kaufhold and Rosenbusch (2013)
‐3.5
‐3
‐2.5
‐2
‐1.5
‐1
‐0.5
0
0.5
80 85 90 95 100
log( m
ortality ha
zard )
Age
Actual
Linear
Logistic
Modeling advanced ages.
3. Mortality law for oldest ages?• Gavrilova & Gavrilov: No deceleration• Ouellette & Bourbeau: Deceleration
Gompertz law (1825): →
∞
Thatcher-Kannisto (1998): →
1(Perks 1932)
Beard (1959): →
Modeling advanced ages.
Log (force of mortality): Beard law (1959)
Source: Richards, Kaufhold and Rosenbusch (2013)
‐3.5
‐3
‐2.5
‐2
‐1.5
‐1
‐0.5
0
0.5
80 85 90 95 100
log( m
ortality ha
zard )
Age
Actual
Linear
Logistic
Beard
Modeling advanced ages.
3. Mortality law for oldest ages?• Gavrilova & Gavrilov: No deceleration• Ouellette & Bourbeau: Deceleration
Gompertz law (1825): →
∞
Thatcher-Kannisto (1998): →
1(Perks 1932)
Beard (1959): →
LESSON 2: Use the Force!
Modeling advanced ages.
Zhiwei Zhu, Zhi Li.: Logistic regression analysis for insured mortality experience study
• Improve use of data and create portfolio-specific tables.• Identify explanatory variables, a.k.a. risk factors.• Multivariate analysis avoids “slicing and dicing”.• Statistical analysis of goodness of fit and significance of
results.• Easy access via stats packages.
Advanced Modeling
How does it work? Parametrise the mortality curve:Let‘s say:
Substitute for : ⇔ ln
, ,…
⋯
1 ⋯
⇔ logit ≔ ln 1 ⋯
Logistic Regression
Linear Regression
MultivariateLinear Regression
Advantages:• Ready access experience analysis tool.• Find and test risk factors:
• Age, gender, selection, smoker status, UW class, time, product type, policy face amount.
Disadvantages:• Restriction on shape.• Grouped data.• Lapses and discontinuities difficult to handle.
LESSON 3: Use all your data!
Logistic Regression
1. Pick any parametric mortality law.Gompertz, Makeham, Perks, Beard, etc.
2. Model the observed lifetime of each individual in continuous time.
3. Maximum Likelihood parameter estimation.
Survival Models
Pick a mortality law: “Makeham−Beard” (Beard, 1959)
1
continuous-time force of mortality.Calculate survival probability:
Survival Models
Likelihood of observing lifetimes in portfolio of lives:
∝
0, 1,
log log
Maximize log with respect to , , , ∈ 1, … , .
Survival Models
Fit a model for each individual by estimating , , ,
…
:
0, 1,
Do the same for , , .
Survival Models
Logarithmic force of mortality by gender
Source: Richard, Kaufhold and Rosenbusch (2013)
Survival Models
‐6
‐5
‐4
‐3
‐2
‐1
0
1
65 70 75 80 85 90 95 100
log ( force of m
ortality )
Age
Actual Female
Actual Male
Model Female
Model Male
Advantages1. Use all data
• Continuous time/age, survivors and deaths
2. Graduate smooth table with flexible shape.3. Automatically weight observations
• Maximum likelihood vs. least mean squares
Disadvantages1. More maths required.
Survival Models
Li, Chan, Li: The CBD Mortality Indices.
• Mortality and longevity risk hedging.
• Non-parametric indices:• London Life and Longevity Market Association: LifeMetrics Index
• Deutsche Börse: Xpect Index
• Proposal: model-based index• Compare to VIX implied volatility indices at CBOE
• Improve information content
• Requirement: Index must be new-data-invariant.
Ideas for trading mortality risk.
M1: The Lee-Carter Model
ln ,
M2: The Renshaw-Haberman Model
ln ,
M3: The Age-Period-Cohort Model
ln ,
Candidate models
M5: The Cairns-Blake-Dowd (CBD) model
ln ,
1 ,̅
M6: The CBD model with a cohort effect
ln ,
1 ,̅
M7: The CBD model with cohort and quadratic effects
ln ,
1 ,̅ ̅
Candidate models
Candidate modelsModel M1: The New‐Data‐Invariant property is NOT satisfied
Estimates of the time‐varying parameters in Model M1Population: English and Welsh malesSample periods: 1950 to 1989, 1994, 1999, 2004, 2009
Candidate modelsModel M5: The New‐Data‐Invariant property is satisfied
Estimates of the time‐varying parameters in Model M5Population: English and Welsh malesSample periods: 1950 to 1989, 1994, 1999, 2004, 2009
The Cairns‐Blake‐Dowd model:
ln ,
1 ,̅
The 1st CBD index ( )Represents the level of the mortality curve (after transformation)A reduction in means an overall mortality improvement.
The 2nd CBD index ( )Represents the slope of the logit‐transformed mortality curveAn increase in means that mortality at younger ages improves more rapidly than that at older ages
Interpretation
Interpretation
Interpretation K1 and K2 risks
Example 1: England & Wales
The centroid moves to the upper‐left
The area becomes larger over time
The portions falling into the lower quadrants are similar in size
Example 2: Canada
Narrower and taller
More tilted
The area falling into the lower‐left‐hand quadrant is smaller
1. Models may break!
2. Use the force!
3. Use all your data!
4. Models can be used to trade risk.
Inspirations & Lessons
54
Communicating Longevity Risk
55
Customers
Senior Management
Chief Risk Officer
Public Policymakers
Identifying Your Stakeholders
Am I going to outlive my savings?
Can we translate risk into opportunity?
How long can people live? Will it be costly?
How do we fund our public pensions?
• Customers do not understand probabilities – make it binary.
• Studies show consumers consistently underestimate their life expectancy.1
• Ask the question: What are the chances you will live to age X or more?
• Make it visual – leverage retirement planning tools that incorporate life expectancy.
• Example: www.troweprice.com/ric
• Focus on the bottom line – reduce spending, invest to remove risk (e.g. annuities).
56
Longevity Risk for Customers
1Source: “Subjective Survival Probabilities and life tables: Evidence from Europe”, Franco Peracchi and Valeria Perotti
• Customer profiles help company leaders understand their oldest customers.
• Longevity Genes Project2
• Align customer needs with product solutions: annuities, income riders, etc.
57
Longevity Risk for Senior Management
2Source: The Longevity Genes Project, Dr. Nir Barzilai, http://www.einstein.yu.edu/centers/aging/longevity‐genes‐project/
• Summarize your changing life tables into life expectancy at key ages.
• Equate ages.
• What if all our customers live 1 year longer?
58
Longevity Risk for Senior ManagementLife expectancy at birth for males3
In 1950 65.4
In 1970 67.0
In 1990 71.9
In 2010 76.4
Current age and age of equivalent mortality 50 years ago for U.S. males4
60 53
70 63
3Source: Human Mortality Database, www.mortality.org4Source: “The Advancing Frontier of Survival: With a Focus on the Future of US Mortality”, James W. Vaupel
• How volatile are adult life expectancies?• Compressing versus shifting.
• Can we look beyond the U.S.?
• Lack of expert consensus on the range of mortality improvement factors.
• 1% (UK Public Pension) to 3% (expert estimates) and beyond (biomedical gerontologists).
• Improvement & population relationship. 59
Longevity Risk for the CRO
• Focus on the total population “Big Picture.” • Improvement by age matters.5
• Reduced mortality under age 35 lower cost.• Reduced mortality over age 35 higher cost.
• Birth rate needs to be presented in parallel.• Importance of retirement age assumption.• Public funding needed to understand new
drivers of old age mortality.• Example: Alzheimer's.
60
Longevity Risk for Public Policymakers
5Source: “Declining Mortality (Increasing Longevity): At What Rate?”, Steve Goss, Office of the Chief Actuary, U.S. S.S.A
Jean-Marc is Vice President, Research and Development at Optimum Re in Dallas, TX.He is responsible for the evaluation of new concepts involving product development or underwriting. His special areas of interest are all things mortality and critical illness both in the US and Canada.Prior to joining Optimum in 1997, Jean-Marc has worked at a number of reinsurer and direct life insurance companies, mostly in the area of product development.
Jean-Marc Fix, FSA, MAAA
Jean-Marc is a frequent speaker at life insurance, underwriting and medical directors meetings and is currently involved with a number of industry activities, including:
SOA Living to Age 100 Symposium Program Committee Joint AAA/SOA team on guaranteed and simplified issue mortality Joint AAA/SOA Underwriting Criteria Team SOA Reinsurance Section Research Team
Jean-Marc received a BA in Mathematics from Whittier College in California
Kai is managing director of Ad Res Advanced Reinsurance Services GmbH in Cologne, Germany. He manages projects in the fields of biometric experience analysis and mortality projection as well as life reinsurance. Prior to founding Ad Res in 2011, Kai worked for Manulife Reinsurance in Toronto and in Cologne. From 2006 to 2011 he was in charge of Manulife’s European life retrocession business, after having had roles in life reinsurance pricing, product development and marketing.
Kai Kaufhold, Aktuar DAV
Kai is a member of the German actuarial association (DAV) and of the IAA Life Section, and enjoys discussing biometric risk analysis and reinsurance where ever he goes. He received an M.Sc. in physics from the University of Cologne.
Andrew is VP of Annuity Product Development at Protective Life Insurance Co.
He is a Fellow of the Society of Actuaries, a Member of the American Academy of Actuaries, a Chartered Enterprise Risk Analyst, a Certified Product Manager, and a Certified Product Marketing Manager.
He holds a B.S. in Mathematics from the University of Texas at Austin (Hook ‘Em).
Andrew Jenkins, FSA, CERA, MAAA