Equity-Based Insurance Guarantees Conference Nov. 11-12, 2019

Chicago, IL

Policyholder Behavior Experience Data

and Modeling

Timothy Paris

SOA Antitrust Compliance Guidelines SOA Presentation Disclaimer

Sponsored by

SOA Equity-Based Insurance Guarantees Conference

Policyholder Behavior Experience Data and ModelingSession 2B

TIMOTHY PARIS, FSA, MAAA

RUARK CONSULTING, LLC

November 11, 2019

1:30 - 3:00pm Central

SOA

Antit

rust

Com

plia

nce

Gui

delin

esAc

tive

part

icip

atio

n in

the

Soci

ety

of A

ctua

ries

is a

n im

port

ant a

spec

t of m

embe

rshi

p. W

hile

the

posi

tive

cont

ribut

ions

of p

rofe

ssio

nal s

ocie

ties

and

asso

ciat

ions

are

wel

l-rec

ogni

zed

and

enco

urag

ed, a

ssoc

iatio

n ac

tiviti

es a

re v

ulne

rabl

e to

clo

se a

ntitr

ust s

crut

iny.

By

thei

r ver

y na

ture

, ass

ocia

tions

brin

g to

geth

er in

dust

ry c

ompe

titor

s an

d ot

her m

arke

t pa

rtic

ipan

ts.

The

Uni

ted

Stat

es a

ntitr

ust l

aws

aim

to p

rote

ct c

onsu

mer

s by

pre

serv

ing

the

free

eco

nom

y an

d pr

ohib

iting

ant

i-com

petit

ive

busi

ness

pra

ctic

es; t

hey

prom

ote

com

petit

ion.

The

re

are

both

stat

e an

d fe

dera

l ant

itrus

t law

s, a

lthou

gh st

ate

antit

rust

law

s cl

osel

y fo

llow

fede

ral l

aw.

The

Sher

man

Act

, is t

hepr

imar

y U

.S. a

ntitr

ust l

aw p

erta

inin

g to

ass

ocia

tion

activ

ities

. T

he S

herm

an A

ct p

rohi

bits

eve

ry c

ontr

act,

com

bina

tion

or c

onsp

iracy

that

pla

ces

an u

nrea

sona

ble

rest

rain

t on

trad

e. T

here

are

, how

ever

, som

e ac

tiviti

es th

at a

re il

lega

l un

der a

ll ci

rcum

stan

ces,

suc

h as

pric

e fix

ing,

mar

ket a

lloca

tion

and

collu

sive

bid

ding

.

Ther

e is

no

safe

har

bor u

nder

the

antit

rust

law

for p

rofe

ssio

nal a

ssoc

iatio

n ac

tiviti

es.

Ther

efor

e, a

ssoc

iatio

n m

eetin

g pa

rtic

ipan

ts sh

ould

refr

ain

from

dis

cuss

ing

any

activ

ity th

at

coul

d po

tent

ially

be

cons

true

d as

hav

ing

an a

nti-c

ompe

titiv

e ef

fect

. Dis

cuss

ions

rela

ting

to p

rodu

ct o

r ser

vice

pric

ing,

mar

ket a

lloca

tions

, mem

bers

hip

rest

rictio

ns, p

rodu

ct

stan

dard

izat

ion

or o

ther

con

ditio

ns o

n tr

ade

coul

d ar

guab

ly b

e pe

rcei

ved

as a

rest

rain

t on

trad

e an

d m

ay e

xpos

e th

e SO

A an

d its

mem

bers

to a

ntitr

ust e

nfor

cem

ent p

roce

dure

s.

Whi

le p

artic

ipat

ing

in a

ll SO

A in

per

son

mee

tings

, web

inar

s, te

leco

nfer

ence

s or

sid

e di

scus

sion

s, y

ou s

houl

d av

oid

disc

ussi

ngco

mpe

titiv

ely

sens

itive

info

rmat

ion

with

com

petit

ors

and

follo

w th

ese

guid

elin

es:

•-D

o no

tdis

cuss

pric

es fo

r ser

vice

s or

pro

duct

s or a

nyth

ing

else

that

mig

ht a

ffect

pric

es

•-D

o no

tdis

cuss

wha

t you

or o

ther

ent

ities

pla

n to

do

in a

par

ticul

ar g

eogr

aphi

c or

pro

duct

mar

kets

or w

ith p

artic

ular

cus

tom

ers.

•-D

o no

tspe

ak o

n be

half

of th

e SO

A or

any

of i

ts c

omm

ittee

s un

less

spec

ifica

lly a

utho

rized

to d

o so

.

•-D

ole

ave

a m

eetin

g w

here

any

ant

icom

petit

ive

pric

ing

or m

arke

t allo

catio

n di

scus

sion

occ

urs.

•-D

oal

ert S

OA

staf

f and

/or l

egal

cou

nsel

to a

ny c

once

rnin

g di

scus

sion

s

•-D

oco

nsul

t with

lega

l cou

nsel

bef

ore

rais

ing

any

mat

ter o

r mak

ing

a st

atem

ent t

hat m

ay in

volv

e co

mpe

titiv

ely

sens

itive

info

rmat

ion.

Adhe

renc

e to

thes

e gu

idel

ines

invo

lves

not

onl

y av

oida

nce

of a

ntitr

ust v

iola

tions

, but

avo

idan

ce o

f beh

avio

r whi

ch m

ight

be

so c

onst

rued

. Th

ese

guid

elin

es o

nly

prov

ide

an o

verv

iew

of

pro

hibi

ted

activ

ities

. SO

A le

gal c

ouns

el re

view

s m

eetin

g ag

enda

and

mat

eria

ls a

s de

emed

app

ropr

iate

and

any

dis

cuss

ion

that

dep

arts

from

the

form

al a

gend

a sh

ould

be

scru

tiniz

ed c

aref

ully

. An

titru

st c

ompl

ianc

e is

eve

ryon

e’s

resp

onsi

bilit

y; h

owev

er, p

leas

e se

ek le

gal c

ouns

el if

you

hav

e an

yqu

estio

ns o

r con

cern

s.

2

Presentation Disclaimer

Presentations are intended for educational purposes only and do not replace independent professional judgment. Statements of fact and opinions expressed are those of the participants individually and, unless expressly stated to the contrary, are not the opinion or position of the

Society of Actuaries, its cosponsors or its committees. The Society of Actuaries does not endorse or approve, and assumes no responsibility for, the content, accuracy or completeness of the

information presented. Attendees should note that the sessions are audio-recorded and may be published in various media, including print, audio and video formats without further notice.

3

Background

4

Why is this important?

• Critical risk element for long-term EBIG-type products, and significant interactions with capital markets risks

• Complex and dynamic data

• Data sparsity, especially at the company level – a credibility problem

• Data is emerging in key areas

• Asset-side folks – demand that your liability-side colleagues demonstrate the robustness of models/assumptions that they provide you

5

Industry studies

Fixed-indexed annuity policyholder behaviorhttps://ruark.co/ruark-releases-2019-fixed-indexed-annuity-study/https://ruark.co/ruark-consulting-releases-2018-fixed-indexed-annuity-mortality-study/

Variable annuity policyholder behaviorhttps://ruark.co/ruark-releases-2019-variable-annuity-study-results/https://ruark.co/ruark-consulting-releases-variable-annuity-mortality-study-results/

6

VM-21 PBR for Variable Annuities

Public redline exposure draft as of April 30, 2019https://naic-cms.org/exposure-drafts

Section 10: Contract Holder Behavior AssumptionsShould examine many factors including cohorts, product features, distribution channels, option values, rationality, static vs dynamicRequired sensitivity testing, with margins inversely related to data credibilityUnless there is clear evidence to the contrary, should be no less conservative than past experience and efficiency should increase over timeWhere direct data is lacking, should look to similar data from other sources/companies

1

2

3

4

7

You and your data

8

0%

35%

7 ormore

6 5 4 3 2 1 0 -1 -2 -3 ormore

Years Remaining in Surrender Charge Period

2008

2016

2018

Your company-level data might indicate some key patterns in surrender behavior

9

GLWB

Surr

ende

r Rat

e

9

0%

30%

7 ormore

6 5 4 3 2 1 0 -1 -2 -3 ormoreYears Remaining in Surrender Charge Period

Surrender rates are lower with living benefits…

10

Surr

ende

r Rat

e

None

GLWB

Hybrid GMIB

10

0%

25%

7 ormore

6 5 4 3 2 1 0 -1 -2 -3 ormore

Surr

ende

r Rat

e

Years Remaining in Surrender Charge Period

GLWB - Withdrawal Behavior

11

…and even lower with income utilization

No prior WDs

Excess WDs

Less than or full WDs

11

…and when guarantees are more valuable

12

GLWB (nominal moneyness basis)

0%

25%

7 ormore

6 5 4 3 2 1 0 -1 -2 -3 ormore

Surr

ende

r Rat

e

Years Remaining in Surrender Charge PeriodITM 50+% ITM 25 - 50% ITM 5 - 25% ATM OTM

12

Dynamic sensitivity has also changed over the years

13

0%

35%

3Q 09 3Q 10 3Q 11 3Q 12 3Q 13 3Q 14 3Q 15 3Q 16 3Q 17 3Q 18

GLWB Shock Lapse

ATM <25% ITM 25%-50% ITM 50%-100% ITM

Surr

ende

r Rat

e

13

How you measure value matters, but company-level credibility is very limited

14

0%

25%

OTM 50+% OTM 25 -50%

OTM 5 -25%

ATM ITM 5 -25%

ITM 25 -50%

ITM 50 -100%

ITM 100%+

Surr

ende

r Rat

eGLWB

Shock -actuarial

Ultimate -actuarial

Shock -nominal

Ultimate -nominal

14

Largest and smallest contracts behave differently

15

0%

20%

7 ormore

6 5 4 3 2 1 0 -1 -2 -3 ormore

Surr

ende

r Rat

e

Years Remaining in Surrender Charge Periodunder 50,000 50,000-100,000 100,000-250,000

250,000-500,000 500,000-1,000,000 >=1,000,000

15

Withdrawals vary by age and tax status

16

QualifiedNon-Qualified

0%

100%

<50 50-59 60-64 65-69 70-79 80+

Freq

uenc

y

Attained Age

GLWB

16

Withdrawal behavior is becoming more efficient

17

0%

50%

Q12008

Q12009

Q12010

Q12011

Q12012

Q12013

Q12014

Q12015

Q12016

Q12017

Q12018

Freq

uenc

y

GLWB

LT Full WDs Full WDs Excess WDs

17

0%

10%

<50 50-59 60-64 65-69 70-79 80-LEA Last Eligible

Annu

itiza

tion

Rate

Hybrid GMIB annuitization rates are low, but company-level credibility is very limited

1818

2012 IAM does not fit VA mortality experience very well

19

0%

150%

0-49 50-54 55-59 60-64 65-69 70-74 75-79 80-84 85-89 90+

% o

f Tab

le

Actual vs. 2012 IAM - Projection G2

Male Count Male Amount Female Count Female Amount Base

19

Evidence of anti-selection for death benefit guarantees

20

0%

200%

1 2 3 4 5 6 7 8 9 10 11+

% o

f Tab

le

Duration

Actual vs. 2012 IAM-G2

LB No LB

20

Results vary over time and between companies

• Each company’s size affects quality of analytical insights and volatility of their own results – a credibility problem

• Composition differences

• Idiosyncratic differences – product features, distribution, closed blocks, etc

• Using only your data, it is very difficult to identify the signal from the noise

21

Building models with your data

22

Modeling and assumptions

• Measuring goodness-of-fit for candidate models

• Testing predictive power on out-of-sample data

• Art + science: choosing, communicating, and ongoing recalibration

23

Goodness of Fit

Predictive Power

24

0

1,000

2,000

3,000

4,000

0 1 2 3 4 5 6 7 8 9 10

Baye

sian

Info

rmat

ion

Crite

rion

(BIC

)

Number of Factors

25

0% 2% 4% 6% 8% 10%

1

2

3

4

5

Coefficient Standard Error

Fact

or

26

27

0.00%

1.00%

2.00%

3.00%

0 1 2 3 4 5 6 7 8 9 10

Avg

Abs A

/E E

rror

Number of Factors

28

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

0% 10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

True

Pos

itive

Rat

e

False Positive Rate29

0

20,000

40,000

60,000

80,000

100,000

120,000

140,000

160,000

180,000

95%

96%

97%

98%

99%

100%

101%

102%

103%

104%

105%

1 2 3 4 5 6 7 8 Actu

als

A/E

Factor Xi

30

-40%

-30%

-20%

-10%

0%

10%

20%

30%

40%

0 10 20 30 40 50 60 70 80 90 100

Actu

al to

Mod

el A

vera

ge

Expected Deciles31

Cost-benefit of industry data

32

Example: variable annuity industry data

• 24 companies

• Seriatim monthly data for policyholder behavior and mortality

• January 2008 through December 2018

• $795 billion ending account value

33

How you measure value matters, and credibility is vastly improved with industry data

34

0%

25%

OTM 50+% OTM 25 -50%

OTM 5 -25%

ATM ITM 5 -25%

ITM 25 -50%

ITM 50 -100%

ITM 100%+

Surr

ende

r Rat

eGLWB

Shock -actuarial

Ultimate -actuarial

Shock -nominal

Ultimate -nominal

34

Industry data shows that GLWB income commencement is highest at issue and after bonuses expire…

35

0%

30%

1 2 3 4 5 6 7 8 9 10 11 12 13

Duration

Freq

uenc

y

35

0%

40%

ITM 100+% ITM 50 -100%

ITM 25 - 50% ITM 5 - 25% ATM OTM 5 - 25% OTM 25+%

Freq

uenc

y

…and that ultimate income commencement is dynamic

36

Dur 11+

Dur 3-10

36

0%

10%

<50 50-59 60-64 65-69 70-79 80-LEA Last Eligible

Annu

itiza

tion

Rate

Industry data shows that hybrid GMIB annuitization rates are backloaded…

3737

…and depend on economic value of other benefits, such as continued income utilization

38

0.0%

2.0%

<95% 95-100% 100-105% 105-110% 110-115% >115%

Annu

itiza

tion

Rate

Ratio of Income PV to Annuitization PV

38

Industry data also makes a better tabular mortality basis…

3939

0%

20%

40%

60%

80%

100%

120%

140%

0-49 50-54 55-59 60-64 65-69 70-74 75-79 80-84 85-89 90+

% o

f Tab

le

Actual vs. Ruark VAM 2015

Male Count Male Amount Female Count Female Amount Base

…and shows how income utilization affects mortality

40

0%

150%

First Year No Prior Withdrawals Prior LT and/or FullWDs only

Any Prior Excess WDs

% o

f Tab

le

Actual vs. RVAM 2015

Qualified

Non-qualified

Total

40

Modeling and assumptions

• Measuring goodness-of-fit for candidate models

• Testing predictive power on out-of-sample data

Using relevant industry data to improve candidate models

• Art + science: choosing, communicating, and ongoing recalibration

41

0

1,000

2,000

3,000

4,000

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Baye

sian

Info

rmat

ion

Crite

rion

(BIC

)

Number of Factors

Company-only

Industry

42

0% 2% 4% 6% 8% 10%

1

2

3

4

5

6

7

8

9

10

Coefficient Standard Error

Fact

or

Company-only

Industry

43

44

0.00%

1.00%

2.00%

3.00%

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Avg

Abs A

/E E

rror

Number of Factors

Industry

Company-only

Customize your model in a credibility-based framework

• Subject matter expertise

• Actuarial judgment

• Quantify the benefits of using relevant industry data

• Ongoing recalibration, so focus on the framework and its sense of range

45

46

0% 2% 4% 6% 8% 10%

1

2

3

4

5

6

7

8

9

10

Coefficient Standard Error

Fact

or

Industry

Customized blend

Company-only

47

0.00%

1.00%

2.00%

3.00%

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Avg

Abs A

/E E

rror

Number of Factors

Company-only

Industry

Customized blend

How much is 1% A/E improvement worth to you?

Suppose 5.00% average annual surrender rates for your block

1% A/E improvement would be 0.05% annually and about 0.60% in present value terms

With 15% annualized market vol, hedge breakage (~2 s.d.) would be 0.18% of notionals

So what are your hedge notionals?

48

Hedge notionals Annualized hedge breakage (~ 2 s.d.)

$100 million $180,000

$1 billion $1,800,000

$10 billion $18,000,000

0.60% * 15% * 2

Cost-benefit of industry data

• Need to customize your model in a credibility-based framework

• Quantify the improvement in goodness-of-fit and predictive power metrics

• Quantify these improvements in financial terms – pricing margins, reserves, hedge breakage

• Quantify the cost to access and use relevant industry data

• Altogether, does this improve your financial risk profile?

Contrast this approach with unlocking ad nauseam

49

More data and/or

relevant industry data

Art + science, subject matter expertise and

actuarial judgment

More statistically justifiable

model factors and

dramatically improved fit

and predictive power

50

Discussion

51