CMI UpdateACA Conference 2015

Katherine Fossett

5 February 2016

CMI Structure

2

Executive Committee

Annuities AssurancesIncome

ProtectionMortality

ProjectionsSAPS

Management Committee

High Age Mortality

5 February 2016

Self Administered Pension Schemes (SAPS)

35 February 2016

SAPS timeline

4

Date Activity

2002 to 2006 Research and consultation

October 2008 “S1” Series tables released (based on data covering 2000-2006)

April 2010 & May 2011 Annual experience updates covering 2001-2008 & 2002-2009

July 2011 Mortality improvements of self-administered pension schemes

May 2012 Analysis of mortality experience by industry classification

May 2012 & April 2013 Annual experience updates covering 2003-2010 & 2004-2011

April - May 2013 Consultation on proposed “S2” Series tables

February 2014 “S2” Series tables released (based on data covering 2004-2011)

July 2014 Annual experience update covering 2005-2012

December 2014 Annual experience update covering 2006-2013

November 2015 Analysis of mortality experience by industry classification

5 February 2016

Industry Analysis – Exposure

55 February 2016

2.8%3.6%

19.2%

8.6%

1.1%

14.7%

5.1%

7.7%

1.6%

9.7%

4.9%

21.1%

Exposure by Industry - Males

1.1%1.4%

8.6%

11.1%

1.1%

15.7%

2.5%

12.8%0.7%

19.1%

2.7%

23.2%

Exposure by Industry - FemalesOil and Gas

Basic Materials

Industrials

Consumer Goods

Health Care

Consumer Services

Utilities

Financials

Technology

Local Authority

Miscellaneous

Not Included inIndustry Analysis

Industry Analysis – Experience – Males

6

80

85

90

95

100

105

110

115

120

80 85 90 95 100 105 110 115 120

Am

ou

nts

10

0A

/E

Lives 100A/E

Oil and Gas

Basic Materials

Industrials

Consumer Goods

Health Care

Consumer Services

Utilities

Financials

Technology

Local Authority

Miscellaneous

All Industries

5 February 2016

Industry Analysis – Experience – Females

7

80

85

90

95

100

105

110

115

120

80 85 90 95 100 105 110 115 120

Am

ou

nts

10

0A

/E

Lives 100A/E

Oil and Gas

Industrials

Consumer Goods

Health Care

Consumer Services

Basic Materials

Utilities

Financials

Technology

Local Authority

Miscellaneous

All Industries

5 February 2016

What is next for SAPS?

• Annual experience update for 2007-2014

– Due for release soon

• Analysis of public sector data

• Mortality Improvements of self-administered pension schemes?

• “S3” Series tables?

– Considering co-graduation

– Awaiting recommendations for high ages

– No immediate plans to begin work on “S3” Series

85 February 2016

High Age Extrapolation of “S2” Series

9

−60%

−40%

−20%

Nil

+20%

+40%

+60%

+80%

60 65 70 75 80 85 90 95 100 105 110

Dif

fere

nc

e in

lo

g μ

x

Age

Difference in log μx SnPMA_X v SnPMA

S1PMA_H

S1PMA_L

S2PMA_H

S2PMA_L

Heavy

Light

• Edge-effects

• No co-graduationHigh age

mortality

5 February 2016

Data Collection – Size of Current Dataset

• Total: 8.7m Male Pensioners and 4.7m Female Pensioners

• Dataset is large but could still be larger

• Continued submissions essential to ensure analyses remain current

105 February 2016

0

200,000

400,000

600,000

800,000

1,000,000

1,200,000

1,400,000

1,600,000

1,800,000

2,000,000

2007 2008 2009 2010 2011 2012 2013 2014

Ex

po

se

d t

o R

isk

(live

s-w

eig

hte

d)

Year

Exposed to Risk by year - 2007-2014 dataset

Male

Female

Lots of data in

earlier years.Later years

incomplete

Data Collection – Submitting Data

• Data submission specifications in SAPS Coding Guide

– Continuation data to be submitted where possible

– Ideally submit data starting from earliest date possible (i.e. include

previous valuation cycles, even if already submitted)

• CMI has reviewed its data handling procedures

• Postcode data has proved insufficient for analysis

– Committee will stop collecting postcode data

115 February 2016

High Age Mortality Working Party

125 February 2016

Background

High Age Mortality Working Party (HAMWP)

• Established Summer 2014

• Members predominantly drawn from CMI investigation committees

HAMWP terms of reference

• Investigate and summarise published research on high age mortality (90+)

• Investigate absolute mortality rates in respect of closing published tables

• Analyse data issues with available data sets (population / portfolio data)

CMI Working Paper 85

• Details analysis to date and possible future work

135 February 2016

Fitting mortality rates at high ages

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

80 90 100 110 120

Mo

rta

lity

Ra

te

Age

ELT16M High gate ELT16M Crude qx

ELT16M Low Gate ELT16M Fitted qx

14

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

80 90 100 110 120M

ort

ality

Ra

teAge

S2PML High gate S2PML Crude qx

S2PML Low Gate S2PML Fitted qx

S2PML Extended qx ELT16M

Observed

deceleration

5 February 2016

Smooth

progression

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

80 90 100 110 120

Mo

rta

lity

Ra

te

Age

S2PML Fitted qx Logistic Regression

Kannisto Regression Gompertz Regression

Academic Contributions

• A range of functional forms

have been proposed for shape

of mortality curve at older ages

• Graph shows impact on

S2PML curve of regressing

common functional forms

• Gompertz: μx = aebx

• Kannisto: μx = aebx/(1+ aebx)

• Logistic: μx = c + aebx/(1+ αebx)

15

Max e100 = 101.3%*

Min e100 = 80.3%*

*relative to S2PML extension (baseline e100 = 2.12)

5 February 2016

Summary of published tables

Table Run in Age Extension Approach Limit mortality rate

ELT16 n/a

Variable-knot spline regression

fitted to 108, used for high age

extensions

m120 = 2 / q120 ~ 1

“S2” Series 95 Cubic spline with constraints μ120 =1 / q120 ~ 0.64

“08” Series 90 Non-linear interpolation μ120 =1 / q120 ~ 0.64

Canadian CPM2014 94

Quartic polynomial to bridge

graduated rate to population

rates at age 103

q114=0.66

US RP-2014Between

75 and 100

Kannisto regression for high ages

75-104 / interpolation between

main regression and high age

Cap on qx of 0.5

165 February 2016

Comparison of published graduations

• S2PML extended

using methodologies

underlying various

industry tables

• Run in age used for

each table applied to

S2PML data

• Very little variation in

e100 for the extensions

shown due to

closeness of qx

values near 100

17

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

70 80 90 100 110 120

Mo

rta

lity

Ra

te

Age

S2PML Fitted qx S2 Extension

08 Extension RP-2014 Extension

ELT15 Extension

“08” extension min e100 = 2.08

RP-2014 extension

max e100 = 2.13

S2PML e100 = 2.12

5 February 2016

Theories on mortality patterns at high ages

• Debate around mortality deceleration vs Gompertz progression at high ages

• A number of authors support deceleration with heterogeneous populations

(frail die young) seen as main cause

• Gavrilov and Gavrilova (2011, 2014) propose no underlying deceleration,

primarily due to:

– Age misreporting

– Aggregation of single year birth cohorts (heterogeneity)

– Studying age-specific probabilities of death rather than force of mortality

– Data recording in older studies was less accurate

185 February 2016

What data issues should be considered?

19

Issue Population Data Portfolio Data

Exposure

estimation

Mid year population

estimates used as proxy for

exposed to risk

Exposed to risk calculated from

data

Death reportingDeath registrations required

within 5 days of death

Late reported / unreported

deaths common

Phantoms

“Phantom” cohorts possible

due to rolling forward of

census data

“Phantom” exposures possible if

deaths not removed from data

set

MigrationAssumption required for

impact on exposures

Tracing overseas deaths difficult,

likely delays in reporting

5 February 2016

Closed cohort mortality

• Considered deaths for cohorts which are essentially extinct (reached their

110th birthday)

• Estimate historical populations (and mortality) from recorded deaths

– Population for age x in calendar y year y = Px,y

– Deaths for age x in calendar y year y = Dx,y

– Pmax,y = D110,y

– Px,y = Px+1, y+1 + Dx,y

• Implied mortality compared against ONS and HMD published mortality

205 February 2016

Closed cohort mortality

21

60

70

80

90

19

61

19

63

19

65

19

67

19

69

19

71

19

73

19

75

19

77

19

79

19

81

19

83

19

85

19

87

19

89

19

91

19

93

19

95

19

97

19

99

20

01

20

03

Ag

e

Year

Extinct Generation mortality / ONS mortality -England & Wales, Males

60

70

80

90

19

61

19

63

19

65

19

67

19

69

19

71

19

73

19

75

19

77

19

79

19

81

19

83

19

85

19

87

19

89

19

91

19

93

19

95

19

97

19

99

20

01

20

03

Ag

e

Year

Extinct Generation mortality / HMD mortality -England & Wales, Males

+4%

+2%

0%

-2%

-4%

5 February 2016

Areas of further investigation – time trends

• No clear evidence for a material change in mortality patterns for lives born

between 1884 and 1898 in England & Wales data.

0.0%

0.1%

0.2%

0.3%

0.4%

0.5%

0.6%

188

4

188

5

188

6

188

7

188

8

188

9

189

0

189

1

189

2

189

3

189

4

189

5

189

6

189

7

189

8

Su

rviv

al P

rob

ab

ilit

y

Year of Birth

Probability of survival to age 105 from age 90

Survival to 105 (F) Survival to 105 (M)

225 February 2016

Summary of findings so far

• Data and modelling issues at very high ages:

– Inconclusive debate on mortality shape at high ages

– Users should consider age misstatement and late reporting issues

– Published closed cohorts mortality appears to be understated

• Impact of different mortality rates derived using different models generally

not material except at very old ages

• Given this and data issues mentioned above, the Working Party feels that

old age extrapolation choices in recent graduated CMI tables were

reasonable

• Second phase of work now in early stages

235 February 2016

Mortality Projections

245 February 2016

CMI Model timeline

Date Model Activity

2004 to 2008 Research and consultation

Nov 2009 CMI_2009 First version of the Model

Nov 2010 CMI_2010 Annual update

Sep 2011 CMI_2011 Annual update – CMI estimate of high age population

Feb 2013 CMI_2012 Annual update – Revised population estimates after 2011 Census

Apr 2013 Consultation on the Model

Sep 2013 CMI_2013 Annual update

Nov 2014 CMI_2014 Annual update – revisions to calibration method

Mar 2015 Consultation on the release date of future updates to the Model

Sep 2015 CMI_2015 Annual update plus paper on recent mortality

Spring 2016 Consultation on the future of the Model

Mar 2017 CMI_2016 First version of revised model

255 February 2016

Estimate ‘long term’

improvements◄◄ Stitch together ►►

Estimate current

improvements

Current CMI Model – very high level overview

Uncertain More uncertain Guess (educated)

26

In the short-term, the best

guide to the likely pace of

mortality improvement is the

most recently observed

experience.

The forces driving mortality

change are themselves

subject to change over time.

Inform assumption choice

less by analysis, more by

subjective judgement.

5 February 2016

Current CMI Model – overview

1. Data

(a) ONS provides raw death and

population data for England &

Wales

(b) Population data at high ages

constructed by CMI (using ONS

methodology)

(c) CMI caps unlikely data points

2. Fit P-spline surface

to estimate smooth

mortality rates

3. Derive annual

rates of mortality

improvements

4. Determine age-

period / cohort

decomposition

5. Projection

(a) Step back two years

for initial rates

(b) Sum separate

projections of age-period

and cohort components

275 February 2016

Recent mortality

285 February 2016

-15,000

-10,000

-5,000

0

5,000

10,000

15,000

20,000

25,000

30,000

35,000

0 4 8 12 16 20 24 28 32 36 40 44 48 52

Week number

Cumulative report deaths by week compared with average over 2005 to 2014

2015

2005

to

2014

Recent mortality

Standardised mortality ratio (2011), England & Wales, and 2000-2011 trend

Note: 2015 data has been estimated based on actual deaths to 31 July.

90%

100%

110%

120%

130%

2000 2005 2010 2015

295 February 2016

Recent mortality improvements

Annual mortality improvements (1976-2015)

Note: 2015 data has been estimated based on actual deaths to 31 July.

−3%

−2%

−1%

Nil

+1%

+2%

+3%

+4%

+5%

+6%

1975 1980 1985 1990 1995 2000 2005 2010 2015

305 February 2016

Recent mortality improvements

Four-year average annualised mortality improvements (1979-2015)

Note: 2015 data has been estimated based on actual deaths to 31 July.

Nil

+1%

+2%

+3%

1975 1980 1985 1990 1995 2000 2005 2010 2015

315 February 2016

Changes between CMI Model versions

Male life expectancy at age 65, male, for different Model versions

Assumptions: S2PMA at 1 January 2007, projected using CMI_20yy_M[1.5%]

21

22

23

2010 2011 2012 2013 2014 2015 2016 2017

CMI_2009 2010

2011 2012

2013 2014

2015

325 February 2016

Financial impact CMI_2015 vs CMI_2014

• Change in cohort expectation of life*:

• Impact of moving from CMI_2012 considerably greater

33

Age Male Female

55 −0.9% −0.9%

65 −1.2% −1.4%

75 −1.8% −1.9%

85 −1.9% −2.0%

* Age exact on 31 December 2015 and S2PxA + CMI_ 2015 vs CMI_2014 (LTR=1.5%).

5 February 2016

Recent mortality – deaths since CMI_2015

345 February 2016

-15,000

-10,000

-5,000

0

5,000

10,000

15,000

20,000

25,000

30,000

35,000

0 4 8 12 16 20 24 28 32 36 40 44 48 52

Week number

Cumulative report deaths by week compared with average over 2005 to 2014

2015

2005

to

2014

Reviewing the CMI Model – Avenues of

investigation

Data

• Correcting for artefacts

• High age mortality

Deterministicprojection

models

• Deal with concerns with the current model

• Consider alternative models

State space modelling

• Able to answer statistically the questions:(a) where are we now?(b) how certain of that are we?

• This would be a fundamental change in methodology

Other

• Guidance on parameterisation

• Coherent modelling of multiple populations

• Cause of death modelling

355 February 2016

What is coming next?

365 February 2016

Things to look out for in 2016

• SAPS annual experience update for 2007-2014

• CMI Model consultation

– Consultation paper during Spring 2016

– Public meetings in April/May 2016

• No CMI Model release during 2016

– CMI_2016 due out in March 2017

• HAMWP second phase working paper

375 February 2016

Questions

The views expressed in this presentation are those of the presenter.

Please send any questions, views or feedback to

info@cmilimited.co.uk

5 February 2016 38

Comments

Continuous Mortality Investigation Limited

Registered in England & Wales (Company number: 8373631)

Registered Office: Staple Inn Hall, High Holborn, London, WC1V 7QJ

Correspondence address: Cheapside House, 138 Cheapside, London, EC2V 6BW, United Kingdom

Email: info@cmilimited.co.uk

Tel: +44 20 7776 3820

Website: www.cmilimited.co.uk (redirects to www.actuaries.org.uk)

Continuous Mortality Investigation Limited (‘the CMI’) is wholly owned by the Institute and Faculty of Actuaries.

Disclaimer: This document has been prepared by and/or on behalf of Continuous Mortality Investigation Limited (CMI). This document does

not constitute advice and should not be relied upon as such. While care has been taken to ensure that it is accurate, up-to-date and useful,

CMI will not accept any legal liability in relation to its contents.

© Continuous Mortality Investigation Limited

5 February 2016