Deloitte. Stochastic Mortality Andrew D Smith [email protected] April 2005.

20
Deloit te. Stochastic Mortality Andrew D Smith [email protected] April 2005

Transcript of Deloitte. Stochastic Mortality Andrew D Smith [email protected] April 2005.

Page 1: Deloitte. Stochastic Mortality Andrew D Smith AndrewDSmith8@Deloitte.co.uk April 2005.

Deloitte.

StochasticMortality

Andrew D [email protected]

April 2005

Page 2: Deloitte. Stochastic Mortality Andrew D Smith AndrewDSmith8@Deloitte.co.uk April 2005.

Deloitte.

AbstractRapid improvements in pensioner mortality have caught many in the insurance industry by surprise. Insurers have strengthened their reserves and increased the prices they charge for annuities.

Classical mortality models are based on a binomial process, where any two individuals die independently of each other. On the other hand, medical breakthroughs and lifestyle changes such as diet and smoking, affect large parts of a population simultaneously, and therefore introduce dependencies between lives. The presentation quantifies the effect such dependencies can have the risk of large losses.

The presentation goes on to develop the concept of a mortality term structure, which describes not only a pattern of deaths but also how insurers might revise their mortality estimates over the term of a contract. The presentation will show the impact of mortality revisions over a one year capital assessment, and on the value of annuity guarantees.

The presentation concludes by extending mortality models to the individual level. Individual mortality models are relevant for assessing the cost of surrender risk on term assurance contracts, or the impact of guaranteed renewal terms.

Page 3: Deloitte. Stochastic Mortality Andrew D Smith AndrewDSmith8@Deloitte.co.uk April 2005.

Deloitte.

How a Life Annuity Works

0

500

1000

65 70 75 80 85 90 95 100

ExpectedMr AdamsMr Barber

Mr Adams and Mr Brown both buy annuities of £1000Payable annually in arrears.Mr Adams dies aged 80Mr Brown dies aged 90

cash

flow

Page 4: Deloitte. Stochastic Mortality Andrew D Smith AndrewDSmith8@Deloitte.co.uk April 2005.

Deloitte.

Reinsurance Contract

Insurer

customers

100 policyholdersOne-off premium aged 65Annual pension until deathExpected payments £2.12m

Reinsurer

Stop loss reinsurance contractPays annuities once aggregate payout exceeds £2.5m

Page 5: Deloitte. Stochastic Mortality Andrew D Smith AndrewDSmith8@Deloitte.co.uk April 2005.

Deloitte.

Distribution Tail:Alternative Models

2000

2500

3000

3500

4000

4500

5000

5500

6000

95% 96% 97% 98% 99% 100%

independent

v1/2 = 1%

v1/2 = 2%

v1/2 = 5%

thou

sand

s

Page 6: Deloitte. Stochastic Mortality Andrew D Smith AndrewDSmith8@Deloitte.co.uk April 2005.

Deloitte.

Modelling Solvency

• To model cash flows– we need to know only the number of deaths

• But to model solvency tests– we need to know what mortality assumptions

might be in use– at a future valuation date– need stochastic assumptions

• Because its no use being solvent (with high probability) in the long term if you’re insolvent next year.

Page 7: Deloitte. Stochastic Mortality Andrew D Smith AndrewDSmith8@Deloitte.co.uk April 2005.

Deloitte.

Regulator’s View

“ we expect firms to consider … how estimates of longevity might change over time thereby affecting the future valuation of realistic liabilities”

FSA insurance regulatory update - March 2005

In theory, also relevant for modelling mortality guarantees, such asguaranteed annuity options. Historically, mortality fluctuations have been as important as interest moves, dragging GAOs into the money.

Page 8: Deloitte. Stochastic Mortality Andrew D Smith AndrewDSmith8@Deloitte.co.uk April 2005.

Deloitte.

Actuaries Debate the Right Assumptions for Today

12

12.5

13

13.5

14

2000 2005 2010 2015 2020 2025 2030

Short CohortMedium CohortLong Cohort

year of annuity price calculationimm

ed

iate

annuit

y v

alu

e @

4.5

% f

or

65

year

old

Page 9: Deloitte. Stochastic Mortality Andrew D Smith AndrewDSmith8@Deloitte.co.uk April 2005.

Deloitte.

Modelling the Assumptions you Might Make Tomorrow

• Underlying idea– proportional hazards

• Distribution F of time of death– Hazard rate = F’(x)/[1-F(x)]– Actuaries call it “force of mortality”

– Transformed Fnew(x) = 1- [1-F(x)]d

– d = deterioration factor

• IID Random deterioration factor

Page 10: Deloitte. Stochastic Mortality Andrew D Smith AndrewDSmith8@Deloitte.co.uk April 2005.

Deloitte.

Mortality Term Structuresnum

ber

of

surv

ivors

0 1 2 3 4 5

01

23

45

01234567

8

9

10

cash flow date

valuationdate

initial valuation based onexpected table

assumptions gradually replaced by reality

Page 11: Deloitte. Stochastic Mortality Andrew D Smith AndrewDSmith8@Deloitte.co.uk April 2005.

Deloitte.

The Olivier-Smith Model

• Notation:– Take a homogeneous cohort– L(t) = number of survivors aged t

– L(s,t) = Es{L(t)}

– Conditional on information at time s

Page 12: Deloitte. Stochastic Mortality Andrew D Smith AndrewDSmith8@Deloitte.co.uk April 2005.

Deloitte.

Olivier-Smith Model (cont)

vv

sGts

tsL

ssL

tsL

ssL

tsL

tsLvts

vvsG

tsL

tsL

tsL

tsL

ssL

ssLssLBinssL

)1,1(

),1(

),1(

),1(

),1(

)1,1(ln),(

),(~)(

)1,1(

),1(

)1,(

),(

)1,1(

),1(),1,1(~),(

1

1

11

)()*,(

binomial model

gamma deterioration factor

Proportional hazards produces biased deathsBias correction so that L(s,t) is a martingale in s

gamma mean 1, variance v

Page 13: Deloitte. Stochastic Mortality Andrew D Smith AndrewDSmith8@Deloitte.co.uk April 2005.

Deloitte.

Individual Mortality

0

1

2

65 70 75 80 85 90 95 100 105 110

Death Indicator

Dual PrevisibleProjection

process At

process Atp

Atp is a finite variation previsible process

Such that At-Atp is a martingale

Slope of Atp, if it exists

Is your own personal hazard rate

Page 14: Deloitte. Stochastic Mortality Andrew D Smith AndrewDSmith8@Deloitte.co.uk April 2005.

Deloitte.

Stochastic Individual Mortality

0

0.5

1

1.5

2

2.5

3

65 70 75 80 85 90 95 100 105 110

suppose my own hazard rateis a stochastic process …

Consider my “state of health”as a Markov process

dual pre

vis

ible

pro

ject

ion

age

Page 15: Deloitte. Stochastic Mortality Andrew D Smith AndrewDSmith8@Deloitte.co.uk April 2005.

Deloitte.

Properties of Individual Mortality

• Consider A∞p

– Atp

• Conditional on information @ t• Either zero (if dead)• Or exponential(1) (if alive)• “A result which will be obvious to

actuaries” (Rogers & Williams)

Page 16: Deloitte. Stochastic Mortality Andrew D Smith AndrewDSmith8@Deloitte.co.uk April 2005.

Deloitte.

Individual Mortality Construction

• Positive stochastic process μt

– function of a Markov vector process

• Integrated value Ct = ∫t μsds

• Exponential RV L

• T (time of death) defined by CT = L

• Atp = min{CT, L}

• Deep question: do all individual (totally inaccessible) models arise in this way?

Page 17: Deloitte. Stochastic Mortality Andrew D Smith AndrewDSmith8@Deloitte.co.uk April 2005.

Deloitte.

Applications of Individual Mortality

• Options on individual mortality– policy selection– options to lapse and re-enter

• Guaranteed Annuity Options• Explaining select mortality table

– common process μt

– different starting populations

Page 18: Deloitte. Stochastic Mortality Andrew D Smith AndrewDSmith8@Deloitte.co.uk April 2005.

Deloitte.

A Final Puzzle

• You have written a portfolio of pensions business– policyholder annuity rate – = better of {9%, open market rate}– mixture of male and female policyholders

• A well-intended regulator enforces unisex annuity pricing– does the GAO liability increase, decrease or

stay the same?

Page 20: Deloitte. Stochastic Mortality Andrew D Smith AndrewDSmith8@Deloitte.co.uk April 2005.

Deloitte.

StochasticMortality

Andrew D [email protected]

April 2005